Skip to main content

GEAIRM 0.1 - Generic Enterprise AI Reference Model

Status: Working Draft - A15 complete; executable formalization complete; governed publication pending; human-readable M01-M17 semantic baseline frozen
Programme: AIPS companion specification
Current slice: GEAIRM-A15 - Release Closure

GEAIRM is the generic enterprise AI reference model. It organizes enterprise AI concepts and relationships while preserving the semantic authority of BPS, AIPS, the KivaNura Agent Ontology, and other applicable reference models and specifications.

Diagram

Machine-Readable Artifact Downloads​

The human-readable GEAIRM specification remains the normative semantic authority. Machine-readable artifacts implement and index that authority; they do not redefine it.

A10 positive-test contract​

GEAIRM-A10 converts the 17 A9 canonical examples into deterministic expected-PASS test fixtures: 15 A7 semantic-graph fixtures plus one A8 Profile fixture and one A8 Extension fixture. The A10 manifest is the authoritative suite index and records stable testcase IDs, A9/M16 provenance, fixture checksums, applicable validation layers, expected outcomes, applicable invariants/A6 coverage and semantic-assessment obligations.

A10 preserves the A9 examples unchanged and creates byte-identical copies under testcases/positive/. Where A6 applies, A10 records SHACL PASS as the expected result for later A12 execution. This expectation does not collapse hybrid-assessment or assessment-only semantics into machine proof. A10 introduces 0 new semantic concepts and 0 new relationships.

A11 negative / semantic-boundary contract​

GEAIRM-A11 introduces 36 deterministic negative/boundary testcases without changing the frozen semantic model:

  • 15 canonical semantic counterexamples project M16-C01 through M16-C15. They are intentionally A7 PASS + A6 PASS, but their declared inference is expected DOES NOT CONFORM under governed semantic assessment. This is a deliberate test of the machine-conformance restraint established by M17-R25.
  • 9 A6 executable-boundary fixtures are minimal Turtle deltas combined with the A4 ontology baseline. Each must produce the declared SHACL violation.
  • 6 A7 structural boundary fixtures are valid JSON documents expected to fail the A7 structural schema.
  • 6 A8 Profile/Extension boundary fixtures are valid JSON documents expected to fail the A8 Profile/Extension schema.

The A11 negative manifest is authoritative for discovery and expected outcomes. A negative testcase is rejected at its declared boundary; A11 does not redefine semantic invalidity as synonymous with JSON Schema or SHACL failure. Rejection of a negative fixture also does not prove that the remainder of a graph is conformant. A11 introduces 0 new semantic concepts, 0 new relationships, and 0 new invariants.

A12 automated conformance-runner contract​

GEAIRM-A12 introduces the deterministic Automated Conformance Runner for the authoritative A10 and A11 testcase manifests. The runner executes 53 manifest-declared testcases and does not treat filesystem directory discovery as authoritative. It verifies fixture checksums and executes the declared machine layers for JSON syntax, A5 context policy, A7/A8 schemas, Turtle parsing, and A6 SHACL where applicable.

A12 preserves the frozen machine-conformance boundary:

  • a testcase runner result of PASS means the observed executable-layer behavior matches the testcase manifest;
  • an A10 positive machine PASS does not establish complete semantic conformance;
  • M16-C01 through M16-C15 remain machine-valid semantic counterexamples whose declared semantic rejection is preserved for governed assessment rather than falsely machine-proven;
  • A6/A7/A8 negative fixtures PASS the test suite only when the runner observes their expected rejection at the declared machine boundary;
  • hybrid-assessment and assessment-only semantics remain NOT_EXECUTED by A12.

The A12 runner is offline and deterministic. For A6, it executes the exact SHACL profile frozen in the current artifact: 24 sh:NodeShape resources using sh:targetNode / sh:targetClass and 34 sh:SPARQLConstraint / sh:select constraints. This is explicitly not a generic SHACL implementation. The runner validates that profile before execution and fails closed if unsupported SHACL constructs are introduced, preventing silent under-validation.

The A12 reference execution records 53/53 testcase expectations matched and 208/208 executable-layer expectations matched, while preserving 41 semantic-assessment declarations, executing 0 semantic assessments, and making 0 complete semantic-conformance claims. A12 introduces 0 new semantic concepts, 0 new relationships, and 0 new invariants.

A13 cross-artifact reconciliation contract​

GEAIRM-A13 introduces deterministic Cross-Artifact Reconciliation across the A1-A12 executable formalization chain. The reconciliation runner treats the A1, A2 and A3 registries as the canonical machine-readable identity/count sources and verifies their projection into A4 RDF/OWL, A5 JSON-LD, A6 SHACL, A7 structural schema and A8 Profile/Extension schema. It then reconciles A9 example indexing, A10 byte-identical positive fixtures and checksums, A11 invariant references and fixture presence, and the A12 runner manifest/reference-report totals and semantic-claim boundary.

A13 also verifies publication/download references and preserves the frozen baseline of 39 GEAIRM-owned elements, 41 canonical relationships, 36 semantic invariants, and 15 external relationship bridges. Reconciliation is therefore a consistency and traceability operation; it does not create semantic authority, strengthen inference, execute governed semantic assessment, or infer GEAIRA architecture.

An A13 reconciliation PASS means that the declared A1-A12 artifacts agree under the deterministic reconciliation checks. It does not establish complete GEAIRM semantic conformance, AIPS Product qualification, governance approval, or architecture approval. A13 introduces 0 new semantic concepts, 0 new relationships, and 0 new invariants.

A14 regression contract​

GEAIRM-A14 establishes the deterministic Regression Suite over the A1-A13 executable baseline. The A14 regression manifest SHA-256 and size pins all 95 predecessor files under static/spec/aips/geairm/0.1/ as they existed at A13. Later-stage files may be added, but any change to a locked predecessor artifact is surfaced as regression drift rather than silently accepted.

A14 re-executes the unchanged A12 conformance runner and A13 reconciliation runner and compares their normalized behavioral results with the A13 reference state. Environment/version metadata is deliberately excluded from the behavioral digest; testcase outcomes, executable-layer outcomes, reconciliation checks, semantic-assessment restraint, and frozen canonical counts remain locked.

A14 also executes four representative negative controls in isolated temporary project copies:

  1. a checksum-only mutation of a positive fixture MUST be detected by A12 as an expectation mismatch;
  2. an unsupported SHACL PropertyShape MUST make A12 fail closed with a runner/configuration error;
  3. drift of the canonical A5 geairm namespace MUST be rejected by A13 reconciliation; and
  4. removal of a required specification publication reference MUST be rejected by A13 reconciliation.

A14 regression PASS means predecessor artifacts remain stable, established A12/A13 executable behavior remains stable, and the declared negative controls are detected. It does not establish complete GEAIRM semantic conformance, AIPS Product qualification, governance approval, or architecture approval. Governed semantic assessment remains unexecuted. A14 introduces 0 new semantic concepts, 0 new relationships, and 0 new invariants.

A15 release-closure contract​

GEAIRM-A15 closes the GEAIRM 0.1 executable release candidate after successful completion of the frozen M01-M17 human-readable semantic baseline and the A1-A14 executable formalization chain. A15 SHA-256 and size locks the complete 100-file A1-A14 predecessor machine-artifact surface, then independently re-runs A12 conformance, A13 reconciliation and A14 regression from the closure candidate.

A15 preserves the frozen machine-readable semantic baseline of 39 GEAIRM-owned elements, 41 canonical relationships, 36 semantic invariants, and 15 external relationship bridges. It introduces 0 new semantic concepts, 0 new relationships, and 0 new invariants.

A15 closure status is deliberately narrower than governed publication:

  • Executable formalization: COMPLETE / CLOSED RELEASE CANDIDATE.
  • Human-readable M01-M17 semantic baseline: COMPLETE / FROZEN.
  • Document status: Working Draft.
  • Governed publication / release promotion: PENDING — separate governance decision.
  • Governed semantic assessment: not replaced by machine execution.
  • Architecture approval: not implied; GEAIRA remains downstream.

An A15 release-closure PASS therefore means that the declared executable candidate is internally intact and reproducible against A1-A14. It does not establish complete semantic conformance, AIPS Product qualification, governance approval, governed final publication, or architecture approval.

With A15 complete, the GEAIRM 0.1 executable-formalization programme is closed. The next programme is GEAIRA 0.1 — Generic Enterprise AI Reference Architecture, which consumes the stabilized GEAIRM semantics and must not retroactively redefine them.


1. GEAIRM-M01 - Semantic Authority & Jurisdiction​

1.1 Purpose of this slice​

The purpose of M01 is to answer one question before GEAIRM defines any substantive AI reference-model semantics:

Which specification or ontology is authoritative for each concept that GEAIRM needs to use?

GEAIRM is intentionally an integrating reference model. A concept MAY appear in GEAIRM even where its semantic definition is owned elsewhere. Inclusion in GEAIRM does not by itself transfer semantic ownership to GEAIRM.

M01 therefore establishes:

  1. semantic authority categories;
  2. the initial concept-authority matrix;
  3. rules for reuse, reference, specialization and semantic-gap creation;
  4. concepts that GEAIRM MUST NOT redefine;
  5. candidate semantic gaps to be resolved by later GEAIRM slices;
  6. the boundary between reference-model semantics and future GEAIRA architectural semantics.

1.2 Source baseline examined​

M01 is grounded in the current AIPS project baseline, especially:

  • docs/ai-product/ontology.md;
  • docs/ai-product/architecture-alignment.md;
  • docs/reference/taxonomy.md;
  • docs/reference/glossary.md;
  • docs/aiprod/* where Product semantic projections are defined;
  • docs/aipds/* where Product deployment semantics are defined;
  • static/spec/aips/0.1/aips-0.1.core.ttl;
  • static/spec/aips/0.1/kivanura-agent-onlology-0.1.ttl;
  • the BPS Product Ontology / Product Intrinsic Nature (PIN) specification supplied alongside AIPS.

The AIPS baseline already defines a substantial AI semantic kernel. GEAIRM MUST build on that kernel rather than create a parallel vocabulary.


2. Semantic Authority Categories​

GEAIRM uses the following authority categories.

CategoryMeaningGEAIRM treatment
EXTERNAL-AUTHORITYThe concept is canonically defined by a specification or ontology outside GEAIRM.Reference or reuse the concept. Do not redefine its intrinsic meaning.
AIPS-AUTHORITYThe concept is canonically defined by AIPS core, AI Product ontology, taxonomy, or governance semantics.Reuse the AIPS concept and identifier where machine-readable identity exists.
AGENT-AUTHORITYThe concept is canonically defined by the KivaNura Agent Ontology.Reuse/reference Agent ontology identity and semantics.
CROSS-DOMAIN-REFERENCEThe concept belongs primarily to another enterprise reference-model domain such as Data or Analytics/Information.Reference the external authority and define only the AI-specific relationship if required.
GEAIRM-CANDIDATEThe current source baseline uses or implies the concept but does not establish a sufficient generic semantic definition for GEAIRM.Treat as a semantic-gap candidate. Later slices must decide whether GEAIRM owns it or references another authority.
GEAIRA-RESERVEDThe concept is architectural rather than reference-model semantic.Defer to GEAIRA; GEAIRM may define the semantic thing but not architectural responsibility, ABBs, planes, registries or realization topology.

A concept MUST NOT be classified as GEAIRM-owned solely because GEAIRM needs to use it.


3. Initial Semantic Authority Matrix​

3.1 BPS / PIN authority​

BPS remains authoritative for generic Product ontology. GEAIRM does not redefine Productness or Product Intrinsic Nature.

ConceptAuthorityGEAIRM treatmentRationale
ProductBPSEXTERNAL-AUTHORITYGeneric Product identity and Product qualification are not AI-reference-model concepts.
Product Intrinsic Nature (PIN)BPSEXTERNAL-AUTHORITYPIN establishes intrinsic Product nature independently of AI-specific semantics.
Value-Bearing EssenceBPSEXTERNAL-AUTHORITYGeneric PIN dimension.
Essential Value CapabilityBPSEXTERNAL-AUTHORITYGeneric PIN dimension.
ConstitutionBPSEXTERNAL-AUTHORITYGeneric PIN dimension.
Identity InvariantBPSEXTERNAL-AUTHORITYGeneric PIN dimension.
Product FamilyBPSEXTERNAL-AUTHORITYAI Product is one Product-family specialization, not a replacement for the BPS family model.
Product KindBPSEXTERNAL-AUTHORITYGeneric Product concept.
Contextual Product RoleBPSEXTERNAL-AUTHORITYBusiness Product and similar roles remain contextual rather than AI-reference-model classifications.
Compound ProductBPS / PINEXTERNAL-AUTHORITYAIPS already treats Compound Product as a broader BPS concept and distinguishes it from Composite derivation.

M01-R01 - Product semantic primacy: GEAIRM MUST NOT create alternative definitions of Product, PIN, Product Family, Product Kind, Contextual Product Role or Compound Product.


3.2 AIPS AI Product and qualification authority​

AIPS is authoritative for AI Product qualification and AI Product-specific intrinsic classifications.

ConceptAuthorityGEAIRM treatment
AI ProductAIPS AI Product OntologyAIPS-AUTHORITY
AI Product VersionAIPS coreAIPS-AUTHORITY
AI Qualification Evidence (AQE)AIPS core / Product ontologyAIPS-AUTHORITY
AI Product Type (APT)AIPSAIPS-AUTHORITY
Primary AI ProductAIPSAIPS-AUTHORITY
Derived AI ProductAIPSAIPS-AUTHORITY
Composite DerivationAIPSAIPS-AUTHORITY
Model ProductAIPS specialization over BPS ProductAIPS-AUTHORITY
AI-enabled ProductAIPS/BPS boundary semanticsAIPS-AUTHORITY for AI-specific distinction
AI AssetAIPS coreAIPS-AUTHORITY
AI ModelAIPS coreAIPS-AUTHORITY
AI DatasetAIPS coreAIPS-AUTHORITY
AI PipelineAIPS coreAIPS-AUTHORITY
AI Evaluation ReportAIPS coreAIPS-AUTHORITY

M01-R02 - Product boundary: GEAIRM MAY describe how AI Products participate in the enterprise AI reference model, but MUST NOT redefine AI Product qualification, AQE, APT, Model Product, AI-enabled Product or Product-family boundaries.


3.3 AIPS AI Primitive and intelligence-semantics authority​

The current AIPS core already defines generic AI concepts that are broader than Product qualification. GEAIRM SHALL reuse them as part of its reference-model vocabulary.

ConceptAuthorityExisting semantic scopeGEAIRM treatment
Artificial Intelligence Primitive (AIPrimitive)AIPS coreComputational capability materially contributing artificial inference, learning, perception, reasoning, generation, planning, optimization, decision formation or intelligent action.AIPS-AUTHORITY; central reused concept.
Learned Model primitiveAIPSAI Primitive typeAIPS-AUTHORITY
Symbolic ReasonerAIPSAI Primitive typeAIPS-AUTHORITY
Knowledge ReasonerAIPSAI Primitive typeAIPS-AUTHORITY
PlannerAIPSAI Primitive typeAIPS-AUTHORITY
Intelligent SearchAIPSAI Primitive typeAIPS-AUTHORITY
OptimizerAIPSAI Primitive typeAIPS-AUTHORITY
Hybrid Intelligence EngineAIPSAI Primitive typeAIPS-AUTHORITY
Intelligence CapabilityAIPS coreWhat intelligence capability is provided.AIPS-AUTHORITY
PerceptionAIPSIntelligence Capability valueAIPS-AUTHORITY
ClassificationAIPSIntelligence Capability valueAIPS-AUTHORITY
DetectionAIPSIntelligence Capability valueAIPS-AUTHORITY
PredictionAIPSIntelligence Capability valueAIPS-AUTHORITY
GenerationAIPSIntelligence Capability valueAIPS-AUTHORITY
Recommendation / RankingAIPSIntelligence Capability valueAIPS-AUTHORITY
ReasoningAIPSIntelligence Capability valueAIPS-AUTHORITY
PlanningAIPSIntelligence Capability valueAIPS-AUTHORITY
OptimizationAIPSIntelligence Capability valueAIPS-AUTHORITY
ControlAIPSIntelligence Capability valueAIPS-AUTHORITY
Execution ModeAIPS coreHow an AI Product is invoked or operated.AIPS-AUTHORITY; later GEAIRM slices may determine whether a generic non-Product relation is needed.
Intelligence EffectAIPS coreWhat an AI Product produces or causes.AIPS-AUTHORITY for the existing controlled vocabulary.
Agency / AutonomyAIPS coreDegree of authority exercised by an AI Product version.AIPS-AUTHORITY for AIPS AR semantics.
EmbodimentAIPS coreRelationship of AI Product behavior to the physical world.AIPS-AUTHORITY for AIPS EM semantics.
Topology ScopeAIPS coreControlled topology scope used in governance.AIPS-AUTHORITY

M01-R03 - AIPS semantic reuse: Where an AIPS identifier already represents a concept needed by GEAIRM, GEAIRM SHOULD reuse or explicitly map to that identifier rather than mint a semantically equivalent GEAIRM term.


3.4 AIPS risk, governance, evidence and lifecycle authority​

ConceptAuthorityGEAIRM treatment
Effective Risk Tier / R0-R4AIPSAIPS-AUTHORITY
Risk Driver AssessmentAIPSAIPS-AUTHORITY
Governance ObligationAIPSAIPS-AUTHORITY
Governance Obligation ProfileAIPSAIPS-AUTHORITY
Product Qualification CurrentnessAIPSAIPS-AUTHORITY
Dependency CurrentnessAIPSAIPS-AUTHORITY
AI Product AssuranceAIPSAIPS-AUTHORITY for AI Product assurance semantics
Audit EvidenceAIPSAIPS-AUTHORITY where used as AIPS governance evidence
Intelligence Behavior & Evolution Contract (IBEC)AIPSAIPS-AUTHORITY
Learning & Inference Contract (LIC)AIPSAIPS-AUTHORITY but deprecated/specialized compatibility concept
Lifecycle StageAIPSAIPS-AUTHORITY for AIPS Product lifecycle classification

M01-R04 - Governance non-duplication: GEAIRM MUST NOT create a second R0-R4 scheme, second Governance Obligation Profile, or competing AIPS Product-assurance taxonomy.

GEAIRM MAY later define broader generic relationships such as an AI capability being evaluated, governed or evidenced, provided those relationships do not redefine AIPS Product-specific governance semantics.


3.5 KivaNura Agent Ontology authority​

The current AIPS package contains a separate KivaNura Agent Ontology explicitly designed as a generic cross-domain ontology and foundational extension point for specifications such as AIPS. GEAIRM therefore treats it as the semantic authority for Agent identity and Agent taxonomy.

ConceptAuthorityGEAIRM treatment
AgentKivaNura Agent OntologyAGENT-AUTHORITY
Individual AgentKivaNura Agent OntologyAGENT-AUTHORITY
Collective AgentKivaNura Agent OntologyAGENT-AUTHORITY
Living AgentKivaNura Agent OntologyAGENT-AUTHORITY
Artificial AgentKivaNura Agent OntologyAGENT-AUTHORITY
Hybrid AgentKivaNura Agent OntologyAGENT-AUTHORITY
Simulated AgentKivaNura Agent OntologyAGENT-AUTHORITY
Software AgentKivaNura Agent OntologyAGENT-AUTHORITY
Robotic AgentKivaNura Agent OntologyAGENT-AUTHORITY
Model-Triggered AgentKivaNura Agent OntologyAGENT-AUTHORITY
Static Orchestrator AgentKivaNura Agent OntologyAGENT-AUTHORITY
Dynamic Reasoning AgentKivaNura Agent OntologyAGENT-AUTHORITY
Autonomous Cognitive AgentKivaNura Agent OntologyAGENT-AUTHORITY
Intelligence StyleKivaNura Agent OntologyAGENT-AUTHORITY
Agent Autonomy LevelKivaNura Agent OntologyAGENT-AUTHORITY
Agent Control ModeKivaNura Agent OntologyAGENT-AUTHORITY
Tool Use PatternKivaNura Agent OntologyAGENT-AUTHORITY
Legal StatusKivaNura Agent OntologyAGENT-AUTHORITY

M01-R05 - Agent semantic primacy: GEAIRM MUST NOT define a competing Agent class or duplicate the existing Agent taxonomy. GEAIRM MAY define relationships between Agent Ontology concepts and AIPS/GEAIRM concepts.

Important distinction: AIPS AgencyAutonomy and Agent Ontology AutonomyLevel are not declared equivalent by the current baseline. GEAIRM MUST NOT collapse them. Their relationship is deferred to M08.


4. Cross-Domain Concepts​

The AIPS source baseline explicitly states that AI Products may consume Data Products, Information Products, knowledge and other AI Products. It also separates AI Product semantics from Data/Information platforms and broader enterprise architecture.

GEAIRM therefore adopts the following jurisdiction rule:

If a concept's primary meaning belongs to Data, Information, Analytics, Product or Agent semantics, GEAIRM references that authority and defines only the AI-specific relationship required by the enterprise AI model.

Initial cross-domain treatment:

ConceptM01 treatmentLater resolution
DataCROSS-DOMAIN-REFERENCEAlign to GEDRM in M07/M17.
Data ProductCROSS-DOMAIN-REFERENCE plus BPS/DPP Product semanticsAlign to GEDRM/BPS/DPP in M07/M12/M17.
InformationCROSS-DOMAIN-REFERENCEAlign to GEARM in M07/M17.
Information ProductCROSS-DOMAIN-REFERENCEAlign to GEARM and applicable Product specification in M07/M12/M17.
Analytics / Analytical OutputCROSS-DOMAIN-REFERENCEAlign to GEARM/GEARA boundary where needed.
ProductEXTERNAL-AUTHORITYBPS.
AgentAGENT-AUTHORITYKivaNura Agent Ontology.

These mappings are jurisdictional placeholders, not new definitions of Data, Information or Analytics.


5. Semantic Gap Candidates​

The current AIPS baseline uses a number of important AI ecosystem terms without establishing a complete generic reference-model definition for each one. M01 does not define these terms. It records them as candidates for later semantic resolution.

Candidate conceptEvidence in current AIPS baselineM01 disposition
Artificial IntelligenceUsed as the domain and in AI Product/AI Primitive definitions, but no standalone generic ArtificialIntelligence class is established in the examined core ontology.GEAIRM-CANDIDATE; resolve in M04.
IntelligenceUsed throughout capability, proposition and effect semantics, but no standalone generic Intelligence entity is established.GEAIRM-CANDIDATE; resolve in M04.
KnowledgeUsed as a dependency/input kind and in Knowledge Reasoner semantics.GEAIRM-CANDIDATE or external authority; resolve in M07.
RuleUsed as a dependency and reasoning input.GEAIRM-CANDIDATE or external authority; resolve in M07.
ContextUsed extensively in purpose, fitness, governance and runtime prose but not established as one generic AI context model.GEAIRM-CANDIDATE; resolve in M07.
Prompt / InstructionUsed as artifact/configuration/evolution terminology; no generic semantic class is established.GEAIRM-CANDIDATE; resolve in M07.
GroundingUsed as an AI input/use role; no generic semantic class is established.GEAIRM-CANDIDATE; resolve in M07.
RetrievalImplied by grounding/search patterns but not established as a generic reference-model concept.GEAIRM-CANDIDATE; resolve in M07.
MemoryUsed in Agent/deployment contexts but no generic enterprise AI Memory concept is established.GEAIRM-CANDIDATE; resolve in M07/M08.
ToolAIPS uses tools as dependency kinds and Agent Ontology defines Tool Use Pattern, but the examined baseline does not define one generic Tool entity class.GEAIRM-CANDIDATE; resolve in M08.
EnvironmentImplied by physical/embedded/agent semantics but not established as one generic environment concept.GEAIRM-CANDIDATE; resolve in M09.
EvaluationAIPS defines evaluation reports, methods, metrics and governance obligations, but a generic evaluation semantic model is not yet established.GEAIRM-CANDIDATE; resolve in M13.
AI EvidenceMultiple evidence types exist, but a single generic evidence super-concept is not established by AIPS beyond PROV-based specializations.Reuse PROV where possible; determine GEAIRM need in M13/M14.
AuthorityAIPS has AuthorityScope, ActionAuthority risk drivers and autonomy semantics; Agent Ontology has control semantics. A single generic AI authority model is not yet established.GEAIRM-CANDIDATE; resolve in M08/M10/M13.

M01-R06 - Gap discipline: A semantic gap candidate MUST NOT become a GEAIRM canonical concept until the responsible later slice demonstrates that (a) the concept is required, (b) no existing authority sufficiently owns it, and (c) its definition does not duplicate another ontology.


6. Terms Requiring Reconciliation Rather Than Redefinition​

Some terms already exist in more than one semantic context. M01 explicitly prevents premature equivalence.

6.1 Autonomy​

  • AIPS defines AgencyAutonomy for AI Product authority behavior: Inform, Recommend, Assist, ActWithApproval, AutonomousAction.
  • The Agent Ontology defines AutonomyLevel: Assisted, SemiAutonomous, Autonomous.

These answer related but different questions. GEAIRM MUST NOT assert equivalence in M01. M08 will define any mapping or relationship.

6.2 Control​

  • AIPS defines Control as an Intelligence Capability.
  • Agent Ontology defines ControlMode as an oversight/control facet.
  • AIPS governance also defines controls and Governance Obligations.

These concepts MUST remain distinct.

6.3 Decision, Plan and Action​

AIPS defines Decision, Plan, DigitalAction and PhysicalAction as Intelligence Effect values. Broader enterprise decision/action semantics also intersect with Analytics and future AI architecture concerns.

GEAIRM MUST preserve the distinction among recommendation, decision, authorization and action. M10 will reconcile the generic semantic relationships without redefining the existing AIPS controlled vocabulary.

6.4 Model​

AIPS distinguishes AI Model, Model Product, AI Primitive and AI Product. These terms MUST NOT be collapsed.

6.5 Composition​

AIPS distinguishes:

  • dependency composition;
  • Derived AI Product classification;
  • Composite derivation topology;
  • broader BPS Compound Product structure.

GEAIRM MUST preserve those distinctions in M11/M12.


7. Namespace and Identifier Rules​

M01 establishes the following identifier policy for later machine-readable formalization.

  1. Existing BPS concepts MUST retain BPS identifiers or explicit BPS mappings.
  2. Existing AIPS concepts MUST retain AIPS identifiers where machine-readable identifiers already exist.
  3. Existing Agent Ontology concepts MUST retain Agent Ontology identifiers.
  4. GEAIRM MUST NOT mint a new identifier solely to provide a local alias for an already authoritative external concept.
  5. A new geairm: identifier MAY be minted only for a concept that passes the semantic-gap test established by M01-R06.
  6. Cross-domain concepts SHOULD be referenced by their authoritative namespace once the relevant external specifications are reconciled.
  7. Human-readable convenience labels MUST NOT imply semantic equivalence where no mapping has been established.

M01-R07 - No shadow ontology: GEAIRM MUST NOT create shadow copies of BPS, AIPS, Agent Ontology, GEDRM or GEARM concepts under a GEAIRM namespace merely for convenience.


8. Reference Model vs Reference Architecture Boundary​

GEAIRM defines semantic concepts and relationships. It does not assign enterprise architectural realization responsibilities.

The following are reserved for GEAIRA rather than GEAIRM:

  • architectural planes;
  • architectural fabrics;
  • architectural capabilities;
  • Architecture Building Blocks (ABBs);
  • registry responsibilities;
  • platform responsibilities;
  • deployment topology;
  • architectural authority boundaries;
  • realization patterns;
  • infrastructure/runtime choices;
  • implementation technologies.

GEAIRM may define the semantic concept that an architecture later realizes. For example, GEAIRM may define or reference Agent, Tool, Evaluation or AIPrimitive; GEAIRA may later define Agent Execution Capability, Tool Registry Responsibility, Evaluation Capability or Primitive Serving ABBs.

M01-R08 - Architecture deferral: GEAIRM MUST NOT make architectural realization structures normative. Those belong to GEAIRA.


9. Product Semantics vs Reference Model Boundary​

GEAIRM is not an AI Product specification.

The following remain AIPS Product concerns:

  • whether a candidate qualifies as an AI Product;
  • AI Qualification Evidence;
  • AI Product Type;
  • Primary/Derived classification;
  • Composite derivation topology;
  • Product Promise / governed Product boundary as defined by applicable Product standards;
  • AIPROD semantic description;
  • AIPDS deployment description;
  • AI Product lifecycle and trust projections.

GEAIRM may model AI Products as participants in the broader enterprise AI semantic landscape, but Product qualification remains externally authoritative.

M01-R09 - Product non-absorption: GEAIRM MUST NOT absorb AIPS Product ontology into the reference model.


10. M01 Canonical Jurisdiction Summary​

The initial semantic ownership model is:

BPS / PIN
owns generic Product semantics
|
+------------------------+
|
AIPS GEAIRM
owns AI Product semantics organizes enterprise AI concepts
owns AI Primitive semantics and their relationships
owns IC / MO / IE / AR / EM |
owns AI risk/governance |
references/reuses
Agent Ontology --------------------------+
owns Agent taxonomy |
owns Agent facets |
|
GEDRM / GEARM and other authorities ------+
own their respective cross-domain meanings

GEAIRM's intended role is therefore:

GEAIRM is the integrating enterprise AI reference model, not the universal semantic owner of every concept that appears in enterprise AI.


11. M01 Normative Rules​

For traceability, the slice establishes the following rules.

RuleRequirement
M01-R01GEAIRM MUST preserve BPS primacy for generic Product semantics.
M01-R02GEAIRM MUST preserve AIPS primacy for AI Product qualification and AI Product-specific classification.
M01-R03GEAIRM SHOULD reuse existing AIPS identifiers for AIPS-owned AI semantics rather than mint equivalents.
M01-R04GEAIRM MUST NOT create competing AIPS risk, governance-obligation or Product-assurance taxonomies.
M01-R05GEAIRM MUST preserve KivaNura Agent Ontology primacy for Agent identity and taxonomy.
M01-R06New GEAIRM concepts require a demonstrated semantic gap and non-duplication check.
M01-R07GEAIRM MUST NOT create shadow copies of externally authoritative concepts under its own namespace.
M01-R08Architectural realization semantics MUST be deferred to GEAIRA.
M01-R09GEAIRM MUST NOT absorb AIPS AI Product ontology into the reference model.
M01-R10Similar terms across authorities MUST NOT be asserted equivalent until a later slice explicitly reconciles them.

12. M01 Exit Criteria​

GEAIRM-M01 is complete when:

  • the authority categories are established;
  • BPS/PIN Product authority is explicit;
  • AIPS-owned AI semantics are inventoried at the reference-model boundary;
  • Agent Ontology authority is explicit;
  • cross-domain Data/Information/Analytics concepts are marked as external rather than silently redefined;
  • semantic-gap candidates are identified without prematurely defining them;
  • namespace/non-duplication rules are established;
  • the GEAIRM/GEAIRA boundary is explicit;
  • later slices have a stable semantic-jurisdiction baseline.

M01 conclusion: these criteria are satisfied by this slice. M02 may now define GEAIRM Purpose, Scope and Positioning without reopening semantic ownership unless a conflict or previously unseen authority is discovered.


13. Forward Work​

The next slice is:

GEAIRM-M02 - Purpose, Scope & Positioning

M02 will define GEAIRM as a reference model, its enterprise scope, explicit non-scope, and its relationship to AIPS, BPS/PIN, GEDRM/GEDRA, GEARM/GEARA, future GEAIRA and HDIP. M02 MUST apply the authority rules established by M01.


14. GEAIRM-M02 - Purpose, Scope & Positioning​

14.1 Purpose of this slice​

M01 established who owns the semantics that GEAIRM must use. M02 establishes what GEAIRM itself is for, where its boundary lies, and how it sits within the wider AIPS, BPS, Data, Analytics, AI Architecture and HDIP landscape.

M02 therefore answers five foundational questions:

  1. What kind of specification is GEAIRM?
  2. What enterprise AI concerns are in scope?
  3. What is explicitly outside GEAIRM's scope?
  4. How does GEAIRM relate to adjacent semantic, Product and architecture specifications?
  5. At what abstraction level does GEAIRM operate?

M02 does not yet define the substantive AI concept model. Those semantics begin in M03-M04 and subsequent slices. M02 establishes the jurisdictional and abstraction boundary within which those later semantics may be defined.


15. Canonical Definition of GEAIRM​

GEAIRM - the Generic Enterprise AI Reference Model - is a technology-neutral reference model that identifies, organizes and relates the generic concepts needed to describe Artificial Intelligence across enterprise contexts, while preserving the semantic authority of the specifications and ontologies that own those concepts.

GEAIRM exists to provide a coherent semantic frame for enterprise AI independent of:

  • a particular AI Product;
  • a particular model family;
  • a particular vendor or foundation model;
  • a particular agent framework;
  • a particular cloud or runtime;
  • a particular industry or legal entity;
  • a particular physical deployment topology.

GEAIRM is therefore concerned with enterprise AI meaning and relationships, not with implementation topology.

A compact statement is:

GEAIRM
= generic enterprise AI concepts
+ their canonical relationships
+ semantic boundaries and invariants
+ explicit mappings to external semantic authorities

M02-R01 - Reference-model purpose: GEAIRM MUST remain a semantic reference model. It MUST NOT become an implementation architecture, deployment architecture or technology catalogue.


16. Meaning of the Name​

16.1 Generic​

Generic means that the model is intended to be reusable across organizations, sectors, technologies and AI mechanisms.

Generic does not mean vague. GEAIRM may define precise relationships and distinctions, but those definitions MUST NOT depend on one organization's platform design or one vendor's terminology.

The genericity requirement means, for example, that GEAIRM may model relationships among intelligence capability, AI Primitive, Agent, Tool, Context and Action without requiring OpenAI, Azure AI, AWS, Google Cloud, LangGraph, PyTorch or another particular technology.

16.2 Enterprise​

Enterprise means that the reference model is capable of describing AI as it participates in a governed organizational ecosystem rather than only as an isolated algorithm or model artifact.

Enterprise scope may therefore include relationships involving:

  • purpose and organizational context;
  • data, information and knowledge dependencies;
  • AI capabilities and mechanisms;
  • people and artificial Agents;
  • tools and environments;
  • recommendations, decisions and actions;
  • Product participation;
  • evaluation, assurance and evidence;
  • lifecycle, provenance and change;
  • risk and governance context.

Enterprise does not mean that every concept is owned by GEAIRM. M01 semantic-authority rules continue to apply.

16.3 AI​

AI is intentionally broader than machine learning.

GEAIRM MUST be capable of representing enterprise AI based on the mechanism-neutral AIPS baseline, including learned, symbolic, knowledge-based, planning, search, optimization, hybrid, agentic and embodied forms where applicable.

GEAIRM MUST NOT define Artificial Intelligence as synonymous with a trained ML model.

16.4 Reference Model​

A Reference Model provides a stable conceptual frame for a domain. It identifies relevant concepts, classifications, roles, states and relationships without prescribing the architecture or technology used to implement them.

The Reference Model answers primarily:

What concepts exist in the enterprise AI domain, what do they mean or reference, and how are they related?

The future GEAIRA Reference Architecture will answer a different question:

What architectural responsibilities, capabilities, building blocks and interactions are required to realize enterprise AI?

M02-R02 - Model-before-architecture: GEAIRM MUST define semantic concepts and relationships independently of the architectural structures that may later realize them in GEAIRA.


17. GEAIRM Scope​

GEAIRM's normative scope is the generic semantic structure of enterprise AI.

Subject to M01 authority rules, GEAIRM may include or reference concepts and relationships concerning the following areas.

17.1 AI intent and enterprise context​

GEAIRM may model the relationship of AI to:

  • purpose;
  • intended outcome;
  • use context;
  • organizational context;
  • consumer or beneficiary context;
  • permitted or constrained behavior;
  • authority context.

GEAIRM does not thereby own generic Product Purpose, Business Context or organizational semantics already defined elsewhere.

17.2 Intelligence capability and mechanism​

GEAIRM may organize relationships among:

  • Intelligence Capability;
  • AI Primitive;
  • learned and non-learned mechanisms;
  • composite and hybrid intelligence;
  • execution mode;
  • intelligence effects.

Existing AIPS identifiers and definitions remain authoritative where M01 assigns AIPS authority.

17.3 Data, information, knowledge and context participation​

GEAIRM may represent how enterprise AI depends on or interacts with:

  • Data;
  • Data Products;
  • Information;
  • Information Products;
  • knowledge;
  • rules;
  • context;
  • grounding;
  • retrieval;
  • memory;
  • prompts or instructions.

The semantic definition of Data, Data Product, Information and Information Product remains with their applicable authorities.

17.4 Agency, autonomy, embodiment and environment​

GEAIRM may represent relationships involving:

  • Agents;
  • agency;
  • autonomy;
  • delegation;
  • oversight;
  • tools;
  • sensing;
  • digital action;
  • physical action;
  • embodiment;
  • operating environment.

Agent identity and taxonomy remain governed by the KivaNura Agent Ontology unless a later approved mapping establishes otherwise.

17.5 Decision and action semantics​

GEAIRM may organize semantic distinctions among AI-generated or AI-mediated:

  • outputs;
  • alerts;
  • recommendations;
  • plans;
  • decisions;
  • authorizations;
  • actions.

GEAIRM MUST preserve the distinction between recommendation, decision and action and MUST align with adjacent Analytics semantics rather than redefine them indiscriminately.

17.6 Evaluation, assurance, governance and evidence​

GEAIRM may model generic relationships involving:

  • evaluation;
  • assurance;
  • evidence;
  • risk context;
  • governance obligations;
  • policy and control relationships;
  • safety and fitness;
  • currentness and trust.

AIPS Product-specific risk, Governance Obligation Profile and assurance semantics remain AIPS-owned.

17.7 Dependency, lifecycle, provenance and evolution​

GEAIRM may model relationships involving:

  • direct and transitive dependency;
  • composition;
  • version dependency;
  • provenance;
  • lifecycle state;
  • semantic change;
  • currentness;
  • evolution;
  • historical classification.

17.8 Product participation​

GEAIRM may represent Products, AI Products, Model Products, Data Products and Information Products as participants in the enterprise AI ecosystem.

Their Product identity and qualification are not established by GEAIRM.

M02-R03 - Broad AI ecosystem scope: GEAIRM MAY describe enterprise AI concepts whether or not the relevant thing qualifies as an AI Product. Productization is not a precondition for participation in the reference model.


18. Explicit Non-Scope​

The following are outside GEAIRM's normative scope unless referenced only to establish a semantic relationship.

18.1 AI Product qualification​

GEAIRM does not determine:

  • whether a candidate is a Product under BPS;
  • whether it qualifies for the AI Product Family;
  • AQE;
  • APT;
  • Primary/Derived classification;
  • Composite derivation topology;
  • Product Promise or AI Product publication state.

These remain AIPS/BPS concerns.

18.2 Architecture design​

GEAIRM does not prescribe:

  • architectural planes;
  • cross-cutting architecture fabrics;
  • architecture capabilities;
  • ABBs;
  • platform responsibilities;
  • registry responsibilities;
  • deployment topology;
  • architecture governance structures.

These belong to GEAIRA or more concrete architecture specifications.

18.3 AI Product deployment​

GEAIRM does not specify how an individual AI Product version is packaged, deployed, served, scaled, monitored or retrained. These are AIPDS and implementation concerns.

18.4 Vendor and technology selection​

GEAIRM does not mandate specific:

  • foundation models;
  • model vendors;
  • cloud providers;
  • agent frameworks;
  • vector databases;
  • AI accelerators;
  • orchestration frameworks;
  • programming languages;
  • deployment products.

18.5 General Data or Analytics semantics​

GEAIRM does not replace GEDRM or GEARM, nor does it redefine Data, Information, Analytics or their complete semantic domains merely because AI consumes or produces them.

18.6 General Product ontology​

GEAIRM does not replace BPS/PIN. In particular, dependency on AI does not by itself change the intrinsic Product Family of another Product. BPS's composition non-inheritance principle remains applicable.

18.7 Law, ethics and regulation as independent domains​

GEAIRM may represent relationships to policy, legal, ethical, safety and regulatory concepts where required by enterprise AI semantics, but it is not itself a complete legal code, ethical theory, regulatory taxonomy or jurisdiction-specific compliance framework.

M02-R04 - Non-scope discipline: Presence of a concept in an enterprise AI scenario MUST NOT be used as justification for GEAIRM to absorb the full semantic domain to which that concept belongs.


19. Relationship to BPS and Product Intrinsic Nature​

BPS provides the generic Product ontology used across Product families. BPS separates Product intrinsic nature from contextual characteristics and allows AI Product to exist as one Product Family without making AI dependencies determinative of every Product's intrinsic identity.

GEAIRM therefore treats BPS/PIN as an orthogonal semantic authority:

BPS / PIN
answers: What is a Product and what is its intrinsic Product nature?

GEAIRM
answers: What concepts participate in enterprise AI and how do they relate?

A Product may participate in GEAIRM without becoming an AI Product. Likewise, a thing represented in GEAIRM need not be a Product at all.

Examples include an AI Primitive, a Tool, an Agent, an evaluation artifact or an environment.

M02-R05 - Product orthogonality: GEAIRM MUST NOT infer Product status or Product Family from mere participation in an AI dependency graph, runtime or interaction.


20. Relationship to AIPS​

GEAIRM is maintained as a companion specification within the AIPS specification family.

AIPS remains authoritative for AI Product qualification and Product-specific AI semantics. GEAIRM broadens the semantic frame beyond qualified AI Products so that the enterprise can describe the wider AI ecosystem in which AI Products, AI Primitives, Agents, Models, tools, knowledge, Data Products, Information Products, people and environments interact.

The relationship is therefore complementary rather than substitutive:

AIPS AI Product semantics
qualify and describe governed AI Products

GEAIRM
organizes the wider enterprise AI concept landscape
and reuses AIPS-owned semantics where applicable

GEAIRM MUST NOT weaken AIPS qualification rules by treating all AI-related assets or capabilities as AI Products.

GEAIRM MAY, however, provide the generic semantic context in which AIPS Product concepts participate.

M02-R06 - AIPS companion boundary: GEAIRM MUST remain compatible with AIPS while remaining semantically broader than AI Product qualification alone.


21. Relationship to the KivaNura Agent Ontology​

The KivaNura Agent Ontology remains the authority for Agent identity and Agent taxonomy established by M01.

GEAIRM's role is to connect Agent concepts into the wider enterprise AI reference model, for example by relating an Agent to:

  • intelligence mechanisms;
  • Intelligence Capabilities;
  • tools;
  • context and memory;
  • delegated authority;
  • recommendations, decisions or actions;
  • other Agents;
  • Products;
  • environments;
  • evaluation and governance evidence.

GEAIRM MUST NOT redefine Agent merely to make these relationships easier to express.

M02-R07 - Agent integration boundary: GEAIRM MAY integrate Agent semantics into enterprise AI relationships but MUST preserve the Agent Ontology as the Agent semantic authority.


22. Relationship to GEDRM and GEDRA​

GEDRM and GEDRA occupy a different domain boundary:

GEDRM
defines/reference-models enterprise Data semantics

GEDRA
defines enterprise Data architectural responsibilities

GEAIRM
defines/reference-models enterprise AI semantics

AI commonly consumes Data and Data Products and may produce Data that later becomes governed or productized. These interactions create relationships between GEAIRM and GEDRM but do not transfer Data semantic authority into GEAIRM.

Likewise, GEAIRM does not define the Data architecture required to source, distribute, govern or productize enterprise Data. Those responsibilities remain with GEDRA and concrete Data architectures.

GEAIRM and GEDRM should therefore be treated as adjacent reference models with explicit cross-domain relationships, not as parent/child models.

M02-R08 - Data boundary: GEAIRM MUST reference rather than absorb authoritative Data semantics and MUST NOT become a substitute for GEDRM or GEDRA.


23. Relationship to GEARM and GEARA​

GEARM and GEARA address enterprise Analytics and Information concerns. GEAIRM and future GEAIRA address enterprise Artificial Intelligence concerns.

The domains overlap in practice. For example:

  • an AI capability may produce an analytical score or recommendation;
  • an Information Product may consume an AI effect;
  • an analytics workflow may invoke an AI Product;
  • an AI Agent may consume an Information Product;
  • a human decision may be informed by both analytical and AI-derived information.

Overlap does not justify semantic collapse.

GEAIRM MUST preserve distinctions such as:

Analytical Output ≠ AI Product
Information Product ≠ AI Product
Recommendation ≠ Decision ≠ Action
Analytics mechanism ≠ AI mechanism merely by implementation label

GEAIRM and GEARM are therefore sibling reference models with intersecting relationships. GEAIRA and GEARA will similarly be sibling reference architectures.

M02-R09 - Analytics boundary: GEAIRM MUST interoperate with GEARM/GEARA without redefining the complete Analytics or Information semantic domain.


24. Relationship to GEAIRA​

GEAIRM is the semantic foundation for the future Generic Enterprise AI Reference Architecture (GEAIRA).

The intended dependency direction is:

GEAIRM
concepts + relationships + invariants
|
v
GEAIRA
architectural responsibilities + capabilities + ABBs + interactions

GEAIRA may define architectural constructs that realize or govern GEAIRM concepts, but it should not need to invent the underlying semantic concepts itself.

Examples:

GEAIRM concept              possible later GEAIRA realization concern
------------------------- -------------------------------------------
Agent Agent execution / coordination responsibility
Tool Tool registration / invocation responsibility
AI Primitive Primitive hosting / serving responsibility
Evaluation Evaluation capability / assurance responsibility
Context Context assembly responsibility
Action Action-control / authorization responsibility

The right-hand column is illustrative only and is not yet normative GEAIRA content.

M02-R10 - Dependency direction: GEAIRA SHOULD depend on the stabilized GEAIRM semantic model; GEAIRM MUST NOT depend on GEAIRA architectural constructs for the meaning of its core concepts.


25. Relationship to HDIP​

HDIP - the Holistic Data, Information and Intelligence Platform - is a more tangible enterprise target conceptual/logical platform architecture.

GEAIRM does not define HDIP. Instead, GEAIRM provides part of the generic semantic foundation that can inform the Intelligence concerns later realized through GEAIRA and specialized within HDIP.

The intended abstraction chain is:

REFERENCE MODELS

GEDRM GEARM GEAIRM
Data Analytics / Enterprise AI
semantics Information semantics
| | |
v v v

REFERENCE ARCHITECTURES

GEDRA GEARA GEAIRA
Data Analytics / AI
architecture Information architecture
\ | /
\ | /
+-----------------+-----------------+
|
v
HDIP
Holistic Data, Information & Intelligence
target conceptual / logical platform
|
v
physical implementation

The visual is conceptual: the domains interact as a graph and are not isolated vertical silos.

HDIP may make enterprise-specific design choices such as platform planes, registries, developer/designer capabilities, Mesh Experience, runtime services and governance mechanisms. Those choices MUST NOT be back-projected into GEAIRM as universal AI semantics.

M02-R11 - HDIP specialization boundary: GEAIRM MUST remain generic and MUST NOT encode HDIP-specific platform organization as intrinsic enterprise AI semantics.


26. Relationship to AIPROD and AIPDS​

GEAIRM, AIPROD and AIPDS operate at different semantic levels.

GEAIRM
generic enterprise AI concept model

AIPROD
authoritative semantic description of a qualified AI Product/version

AIPDS
mechanism-neutral operational/deployment description of that AI Product/version

GEAIRM may provide concepts referenced by AIPROD or AIPDS, but it does not replace either specification.

AIPROD remains concerned with the qualified AI Product/version and its authoritative Product semantics. AIPDS remains concerned with how that Product/version operates and is deployed.

M02-R12 - Product-instance boundary: GEAIRM MUST NOT use generic reference-model semantics to override the authoritative Product/version semantics recorded by AIPROD or the deployment semantics recorded by AIPDS.


27. Abstraction and Realization Levels​

GEAIRM participates in a larger standards and architecture landscape with several distinct abstraction levels.

LevelPrimary concernExamples
Generic Product ontologyWhat is a Product?BPS / PIN
Domain reference modelWhat concepts exist and how do they relate?GEDRM, GEARM, GEAIRM
Product-family specializationWhen does a Product qualify for a specialized Product family?AIPS AI Product semantics, DPP
Reference architectureWhat architectural responsibilities and structures realize the domain?GEDRA, GEARA, future GEAIRA
Enterprise target logical architectureHow does a particular enterprise organize realizable platform capabilities?HDIP
Product deployment specificationHow does a particular Product/version operate?AIPDS
Physical implementationWhich technologies, products and deployments implement the architecture?enterprise-specific technology choices

These levels may cross-reference one another, but they MUST NOT be treated as interchangeable.

M02-R13 - Abstraction preservation: GEAIRM relationships MUST be expressed at reference-model level and MUST NOT silently encode assumptions that belong only to Product qualification, architecture, deployment or physical implementation.


28. Technology and Mechanism Neutrality​

GEAIRM inherits the mechanism-neutral direction already established by AIPS.

The model MUST be capable of representing, where semantically applicable:

  • learned-model intelligence;
  • symbolic reasoning;
  • knowledge reasoning;
  • planning;
  • intelligent search;
  • optimization;
  • hybrid intelligence;
  • agentic systems;
  • multi-agent systems;
  • digital action;
  • embodied and physical AI.

No concept may be classified as non-AI merely because it lacks a trained ML model where the authoritative AIPS semantics allow another qualifying AI Primitive or intelligence mechanism.

Conversely, use of an ML model does not by itself establish that the containing Product is an AI Product.

M02-R14 - Mechanism neutrality: GEAIRM MUST remain neutral among qualifying AI mechanisms and MUST NOT use model as a synonym for AI, AI Primitive, AI capability or AI Product.


29. Enterprise Neutrality and Specialization​

GEAIRM is designed to be specialized by enterprises without losing its generic semantic core.

An enterprise MAY:

  • map internal vocabulary to GEAIRM concepts;
  • apply Profiles once the Profile mechanism is defined later;
  • add organization-specific concepts through governed extensions;
  • map platform capabilities or implementation technologies to GEAIRM concepts through architecture specifications;
  • define stricter governance or applicability rules.

An enterprise specialization MUST NOT:

  • redefine externally authoritative concepts while retaining the same identifier;
  • weaken a normative GEAIRM distinction;
  • assert equivalence without semantic justification;
  • confuse a local architecture realization with the generic reference-model concept.

Formal Profile and Extension semantics are deferred to later normative/executable work.


30. M02 Canonical Positioning Summary​

GEAIRM's position can be summarized as follows:

                         BPS / PIN
generic Product ontology
|
Product semantics
|
+---------+---------+
| |
AIPS GEAIRM
AI Product qualification enterprise AI
and Product semantics reference model
| |
| reuses/references AIPS,
| Agent Ontology, GEDRM,
| GEARM and other authorities
| |
+---------+---------+
|
v
GEAIRA
enterprise AI reference architecture
|
v
HDIP
enterprise target logical platform architecture
|
v
physical implementation

This diagram expresses dependency and positioning, not ownership transfer. GEAIRM remains governed by the semantic-authority rules established in M01.

The canonical short-form responsibility split is:

BPS / PIN
What is a Product?

AIPS AI Product semantics
When does a Product qualify as an AI Product and how is it classified?

GEAIRM
What generic concepts participate in enterprise AI and how do they relate?

GEAIRA
What architectural responsibilities are required to realize enterprise AI?

AIPDS
How does a particular qualified AI Product/version operate and deploy?

HDIP
How does the enterprise organize tangible Data, Information and Intelligence
platform capabilities to realize its target architecture?

31. M02 Normative Rules​

For traceability, M02 establishes the following rules.

RuleRequirement
M02-R01GEAIRM MUST remain a semantic reference model and MUST NOT become an implementation or deployment architecture.
M02-R02GEAIRM MUST define semantic concepts independently of the architectural constructs that may later realize them in GEAIRA.
M02-R03GEAIRM MAY describe enterprise AI concepts whether or not the relevant thing qualifies as an AI Product.
M02-R04GEAIRM MUST NOT absorb an adjacent semantic domain merely because its concepts participate in AI scenarios.
M02-R05GEAIRM MUST NOT infer Product status or Product Family from participation in an AI dependency graph, runtime or interaction.
M02-R06GEAIRM MUST remain compatible with AIPS while remaining semantically broader than AI Product qualification alone.
M02-R07GEAIRM MAY integrate Agent semantics but MUST preserve the KivaNura Agent Ontology as Agent semantic authority.
M02-R08GEAIRM MUST reference rather than absorb authoritative Data semantics and MUST NOT replace GEDRM or GEDRA.
M02-R09GEAIRM MUST interoperate with GEARM/GEARA without redefining the complete Analytics or Information semantic domain.
M02-R10GEAIRA SHOULD depend on stabilized GEAIRM semantics; GEAIRM MUST NOT depend on GEAIRA constructs for the meaning of its core concepts.
M02-R11GEAIRM MUST remain generic and MUST NOT encode HDIP-specific platform organization as intrinsic enterprise AI semantics.
M02-R12GEAIRM MUST NOT override authoritative AIPROD Product/version semantics or AIPDS deployment semantics.
M02-R13GEAIRM MUST preserve reference-model abstraction and MUST NOT silently encode Product, architecture, deployment or physical-implementation assumptions.
M02-R14GEAIRM MUST remain mechanism-neutral and MUST NOT use model as a synonym for AI, AI Primitive, AI capability or AI Product.

32. M02 Exit Criteria​

GEAIRM-M02 is complete when:

  • GEAIRM has a canonical definition;
  • Generic, Enterprise, AI and Reference Model have explicit scope meanings;
  • normative scope is established;
  • explicit non-scope is established;
  • the relationship to BPS/PIN is explicit;
  • the relationship to AIPS and the Agent Ontology is explicit;
  • the relationship to GEDRM/GEDRA and GEARM/GEARA is explicit;
  • the dependency direction to GEAIRA is explicit;
  • the relationship to HDIP is explicit;
  • the relationship to AIPROD/AIPDS is explicit;
  • abstraction levels are separated;
  • technology/mechanism neutrality is explicit;
  • all positioning statements preserve M01 semantic authority.

M02 conclusion: these criteria are satisfied by this slice. GEAIRM now has a stable purpose, scope and positioning boundary within which M03 can define the GEAIRM reference-model metamodel and modelling rules.


33. Forward Work​

The next slice is:

GEAIRM-M03 - Reference-Model Metamodel & Modelling Rules

M03 will define the modelling primitives that GEAIRM itself may use - including Concept, Classification, Relationship, Role, State, Context and external semantic reference - and will define the rules for expressing GEAIRM-owned concepts versus referenced concepts without weakening M01/M02 boundaries.


34. GEAIRM-M03 - Reference-Model Metamodel & Modelling Rules​

34.1 Purpose of this slice​

M01 established semantic jurisdiction. M02 established GEAIRM purpose, scope and positioning. M03 now defines the modelling language that GEAIRM itself may use to express the enterprise AI reference model.

M03 answers four questions:

  1. What kinds of model element may appear in GEAIRM?
  2. How are concepts, classifications, roles, states, relationships and contextual qualifiers distinguished?
  3. How does GEAIRM represent externally authoritative concepts without creating shadow definitions?
  4. What inference and identity rules apply before substantive AI semantics are introduced in M04 and later slices?

M03 is deliberately domain-light. It defines the reference-model metamodel and modelling discipline; it does not yet define Artificial Intelligence, Intelligence or the substantive AI concept domains.

M03-R01 - Metamodel/domain separation: GEAIRM metamodel constructs MUST be kept distinct from the enterprise AI domain concepts represented using those constructs.


35. Metamodel Design Principles​

The GEAIRM metamodel is governed by the following principles.

35.1 Semantic minimality​

GEAIRM SHOULD introduce only the modelling constructs required to express stable reference-model semantics. It MUST NOT create a large upper ontology merely to make the model self-contained.

35.2 Authority preservation​

Every represented concept MUST retain its semantic authority as established by M01. Representation inside GEAIRM does not transfer ownership.

35.3 Explicit relationship semantics​

Relationships MUST be explicitly defined. Graph proximity, similar labels, visual grouping or shared participation in a scenario MUST NOT be treated as semantic equivalence.

35.4 Context without identity collapse​

A thing may play different Roles, occupy different States or participate under different Context qualifiers without thereby becoming a different intrinsic Concept.

35.5 Technology neutrality​

The metamodel MUST remain usable for learned, symbolic, knowledge-based, planning, search, optimization, hybrid, agentic and embodied AI without privileging a vendor, runtime or model family.

35.6 Machine-formalization readiness​

Human-readable constructs defined by M03 SHOULD be capable of later projection into the A1-A8 machine-readable programme without requiring a different conceptual model.


36. GEAIRM Modelling Levels​

GEAIRM distinguishes three modelling levels.

LevelMeaningExamples
L0 - Reference-model metamodelThe constructs used to build GEAIRM.Concept, Classification, Role, State, Relationship Type, Semantic Assertion, Modelling Context, External Semantic Reference, Invariant.
L1 - Enterprise AI reference modelGeneric domain concepts and relationships represented by GEAIRM.AIPS AIPrimitive, Agent Ontology Agent, and later GEAIRM-owned concepts that pass M01-R06.
L2 - Scenario or implementation instanceA concrete Product, Agent, model, deployment, organization or use-case instance.A specific fraud model version, a named enterprise Agent, a particular Product version or a particular deployment.

A normative GEAIRM concept belongs at L1 unless it is explicitly part of the L0 metamodel. L2 examples MAY be used to explain or test GEAIRM but MUST NOT be mistaken for canonical L1 concepts.

M03-R02 - Level discipline: GEAIRM MUST NOT infer an L1 generic concept from an L2 implementation detail, nor treat an L0 modelling construct as if it were an enterprise AI domain concept.


37. Core GEAIRM Metamodel Constructs​

37.1 Reference Model Element​

Reference Model Element is the abstract authoring category for anything that GEAIRM records as part of the model or its semantic governance.

A Reference Model Element may be:

  • a Concept;
  • a Classification;
  • a Role;
  • a State;
  • a Relationship Type;
  • a Semantic Assertion;
  • a Modelling Context;
  • an External Semantic Reference;
  • a Semantic Mapping;
  • an Evidence Reference;
  • an Invariant.

Reference Model Element is a metamodel category only. It is not an enterprise AI concept and MUST NOT be used as a semantic superclass for domain concepts in later ontology artifacts unless the executable formalization explicitly models metamodel resources separately from domain classes.

37.2 Concept​

A Concept represents a generic thing, capability, phenomenon, artifact, participant or other semantic entity that is meaningful in the enterprise AI domain.

A Concept may be GEAIRM-owned or externally authoritative.

A candidate SHOULD be modelled as a Concept when:

  1. it has a stable meaning independent of one scenario;
  2. it can participate in one or more relationships;
  3. it is not merely a temporary Role, State or contextual label; and
  4. its semantic authority can be identified.

Examples of externally authoritative Concepts already known to GEAIRM include AIPS AIPrimitive and Agent Ontology Agent. Their appearance in GEAIRM does not make GEAIRM their semantic owner.

M03-R03 - Concept criterion: A term MUST NOT be promoted to a canonical GEAIRM Concept merely because it appears frequently in AI prose or implementation terminology.

37.3 Classification​

A Classification is a governed categorization applied to a Concept or other eligible model element according to explicit criteria.

Classification answers a question such as:

Which governed category applies to this thing under classification scheme S?

A Classification is not automatically:

  • an intrinsic type;
  • a superclass/subclass relation;
  • a Product Family;
  • a Role;
  • a lifecycle State.

Existing AIPS controlled classifications remain AIPS-owned where established by M01.

M03-R04 - Classification discipline: Applying a Classification MUST NOT by itself change the identity or semantic authority of the classified Concept.

37.4 Role​

A Role is a contextual function, participation mode or responsibility that a Concept plays within a defined relationship or context.

A Role is relational and contextual. The same Concept may play multiple Roles simultaneously or at different times.

Examples of the modelling idea include producer/consumer-like participation, tool-user/tool-provider participation or evaluator/evaluated participation where later slices establish such domain relationships.

M03 does not define those domain Roles; it defines how Role is used as a metamodel construct.

M03-R05 - Role non-intrinsicality: A Role MUST NOT be treated as an intrinsic type of the Concept that plays it unless an authoritative semantic specification explicitly defines that identity relationship.

37.5 State​

A State represents a condition of a Concept or relationship at a relevant point or interval in time.

State is intended for semantics such as lifecycle, currentness, availability, activation or other condition-based classification once later slices define or reference the relevant domain states.

A State differs from a Concept type: moving from one State to another normally does not imply that the thing has become a different intrinsic kind.

M03-R06 - State non-taxonomy: A State transition MUST NOT be represented as taxonomic specialization unless the applicable semantic authority explicitly defines it that way.

37.6 Modelling Context​

Modelling Context is an L0 qualifier used to scope the interpretation of an assertion.

It may qualify, for example:

  • organizational scope;
  • jurisdiction;
  • use context;
  • temporal interval;
  • authority scope;
  • consumer scope;
  • scenario scope.

This L0 construct does not settle the M01 semantic-gap candidate Context as an enterprise AI domain Concept. The generic AI-domain meaning of Context remains deferred to M07.

M03-R07 - Context-level distinction: The L0 Modelling Context qualifier MUST NOT be used to pre-empt or silently define the L1 enterprise AI concept Context reserved for M07 analysis.

37.7 Relationship Type​

A Relationship Type defines a permitted semantic connection between model elements.

Every canonical Relationship Type MUST define, at minimum:

  • a preferred label;
  • intended meaning;
  • semantic authority;
  • permitted subject kind(s);
  • permitted object kind(s);
  • directionality;
  • whether symmetry, transitivity, inverse semantics or other inference behavior is explicitly authorized;
  • any important non-implications.

Unless explicitly declared otherwise, a Relationship Type is treated as:

  • directional;
  • non-symmetric;
  • non-transitive;
  • non-functional;
  • without an implied inverse.

M03-R08 - Relationship explicitness: GEAIRM MUST NOT infer symmetry, transitivity, inverse, equivalence, containment or ownership transfer from a Relationship Type unless that behavior is explicitly authorized.

37.8 Semantic Assertion​

A Semantic Assertion is a statement that a subject model element participates in a Relationship Type with an object model element, optionally qualified by Modelling Context, State, provenance or evidence reference.

A compact form is:

Semantic Assertion
= Subject
+ Relationship Type
+ Object
+ optional Context / State qualifiers
+ semantic authority / provenance

A Semantic Assertion may be normative or informative depending on its status in the specification.

M03-R09 - Assertion traceability: Every normative Semantic Assertion SHOULD be traceable to the slice or external authority that establishes its meaning.

37.9 External Semantic Reference​

An External Semantic Reference records that a model element or assertion depends on a concept whose semantic authority is outside GEAIRM.

An External Semantic Reference SHOULD record, where available:

  • semantic authority name;
  • authoritative identifier or IRI;
  • authoritative preferred label;
  • specification/ontology version or version policy;
  • GEAIRM representation disposition;
  • any approved semantic mapping.

An External Semantic Reference is not a local redefinition.

M03-R10 - External-reference fidelity: GEAIRM MUST preserve the authoritative identity of an externally owned concept where that identity exists and MUST NOT substitute a local GEAIRM identity merely for convenience.

37.10 Semantic Mapping​

A Semantic Mapping records an explicit semantic relationship between two terms or concepts, usually across authorities.

Mappings may later use relations such as:

  • equivalent to;
  • specializes;
  • broader than;
  • narrower than;
  • related to.

These mapping categories are not interchangeable. In particular, equivalent to is a strong assertion and MUST NOT be inferred from similar names or overlapping use cases.

M03-R11 - Equivalence discipline: Semantic equivalence MUST be asserted only when the relevant authorities and definitions justify substitutability of meaning; lexical similarity is insufficient.

37.11 Evidence Reference​

An Evidence Reference is an L0 pointer from a model assertion or modelling decision to supporting specification text, ontology material, test evidence or other authoritative source.

This is a metamodel traceability construct. It does not establish the future L1 domain concept AI Evidence, which remains subject to M13/M14 analysis under M01.

M03-R12 - Evidence-level distinction: L0 Evidence Reference MUST NOT be treated as a canonical enterprise AI Evidence concept.

37.12 Invariant​

An Invariant is a normative semantic condition that must remain true across conforming interpretations of GEAIRM.

An Invariant may constrain:

  • identity distinctions;
  • allowed or forbidden relationships;
  • classification behavior;
  • semantic ownership;
  • inference behavior;
  • context-sensitive interpretation.

M15 will consolidate the canonical semantic invariant catalogue. M03 establishes only the metamodel notion and the modelling rules required before then.


38. Representation Disposition vs Semantic Authority​

M01 defines semantic authority categories. M03 adds a separate concept: representation disposition.

Semantic authority answers:

Who owns the meaning?

Representation disposition answers:

How does GEAIRM represent or use that meaning?

The following dispositions are permitted during the normative programme.

DispositionMeaning
OWNEDGEAIRM owns the canonical definition because the concept passed the M01 semantic-gap test.
REUSEDGEAIRM uses an externally authoritative concept as a first-class participant while preserving its external identity and authority.
REFERENCEDGEAIRM points to an external concept without importing or redefining it as a local canonical concept.
MAPPEDGEAIRM records an explicit semantic mapping between concepts or authorities.
CANDIDATEThe term is under semantic-gap analysis and is not yet canonical.
RESERVEDThe concern belongs outside GEAIRM, for example to GEAIRA architecture or AIPDS deployment semantics.

These dispositions do not supersede M01 authority categories.

For example:

AIPS AIPrimitive
semantic authority: AIPS
GEAIRM disposition: REUSED

Agent Ontology Agent
semantic authority: KivaNura Agent Ontology
GEAIRM disposition: REUSED

Grounding (before M07 resolution)
semantic authority: unresolved
GEAIRM disposition: CANDIDATE

Agent Execution Capability
semantic authority: future GEAIRA architecture
GEAIRM disposition: RESERVED

M03-R13 - Authority/disposition separation: Representation disposition MUST NOT be used to imply transfer of semantic authority.


39. Taxonomy, Classification, Role and State​

GEAIRM MUST distinguish four different modelling questions.

Modelling formQuestion answeredTypical semantic effect
Taxonomic specializationWhat more-specific kind of Concept is this?Intrinsic/type-level relation where authoritative semantics justify it.
ClassificationWhich governed category applies?Adds a classification without necessarily changing identity.
RoleWhat function does this Concept play here?Contextual participation.
StateWhat condition is this Concept currently in?Temporal or lifecycle condition.

The four MUST NOT be collapsed for modelling convenience.

Illustrative examples:

  • a Product being classified R3 is not thereby a subtype called R3Product;
  • an Agent playing an evaluator Role does not become intrinsically a new kind of Agent unless the Agent authority defines such a type;
  • a lifecycle State does not create a new intrinsic Product or AI Concept identity;
  • a true taxonomic specialization requires semantic justification from the relevant authority.

M03-R14 - Facet separation: Taxonomy, Classification, Role and State MUST remain semantically distinct unless an authoritative specification explicitly relates them.


40. Scoped and Contextual Assertions​

Enterprise AI semantics are frequently scoped. A statement that is true in one organizational, temporal, authority or use context may not be universally true.

GEAIRM therefore permits assertions to carry explicit qualifiers.

A conceptual assertion form is:

Subject
-- Relationship --> Object
[within Modelling Context C]
[while in State S]
[according to Authority A]

A qualifier narrows the applicability of the assertion. It does not automatically modify the intrinsic identity of Subject or Object.

Examples of later potential scoping include:

  • action authority valid only within a defined task context;
  • an Agent operating under a particular oversight mode;
  • an evaluation result valid for a particular version and evaluation context;
  • a dependency valid only for a specific Product version.

M03 establishes the scoping mechanism only; later slices own the substantive semantics.

M03-R15 - Scoped assertion discipline: A context-qualified assertion MUST NOT be generalized beyond its declared scope without explicit semantic justification.


41. Relationship and Inference Discipline​

GEAIRM is a semantic graph, but a graph MUST NOT be treated as an unrestricted inference engine.

41.1 No inference from visual layout​

Visual adjacency, containment in a diagram, arrow direction chosen for readability, color or grouping MUST NOT create semantics that are absent from the normative relationship definition.

41.2 No inheritance through dependency​

If A depends on B, A does not thereby inherit B's type, Product Family, authority, risk tier, role or other classifications.

This is consistent with the BPS composition non-inheritance principle and the M01 Product boundary.

41.3 No authority transfer through relationship​

If a GEAIRM-owned Concept relates to an AIPS-owned Concept, that relationship does not transfer ownership of the AIPS Concept to GEAIRM.

41.4 No Product inference​

Participation in AI execution, dependency, reasoning, agent activity or an AI graph does not make a thing an AI Product. M02-R05 remains controlling.

41.5 No architecture inference​

A semantic relationship does not determine where a capability must be deployed, which platform must provide it or which ABB realizes it. Those are GEAIRA or implementation concerns.

M03-R16 - Inference restraint: Only inference behavior explicitly authorized by normative GEAIRM or an incorporated authoritative semantic source may be relied upon for conformance or interpretation.


42. External Authority Mapping Rules​

When GEAIRM references or maps an external concept, the following rules apply.

  1. The authoritative concept identity SHOULD be preserved where available.
  2. The semantic authority MUST be recorded.
  3. The mapping relation MUST be explicit.
  4. Equivalent meaning MUST NOT be claimed merely to simplify diagrams or code.
  5. Version-sensitive external semantics SHOULD record the applicable authority version or version policy.
  6. A local convenience label MAY be displayed, but MUST NOT masquerade as a new canonical GEAIRM identity.
  7. If an external authority changes incompatibly, the mapping MUST be reassessed rather than silently carried forward.
  8. Where no authoritative identifier exists, the human-readable authority reference MUST be sufficiently precise to identify the source definition.

This gives later executable artifacts a clean route to machine-visible semantic ownership.


43. Identity, Labels and Namespace Discipline​

GEAIRM distinguishes identity from label.

A preferred label is human-readable text. It is not, by itself, semantic identity.

Two terms may share a label and remain different concepts. Two concepts may use different labels and still later be mapped as equivalent if authoritative definitions justify that mapping.

During M01-M17:

  • externally authoritative identifiers MUST be preserved where available;
  • GEAIRM-owned canonical identifiers MAY be assigned only after a concept passes M01-R06;
  • working labels for CANDIDATE concepts MUST NOT be treated as stable identifiers;
  • aliases MAY aid readability but MUST NOT imply equivalence;
  • namespace allocation MUST avoid shadow copies of external terms.

The executable A1 Concept Registry will later freeze machine identifiers for canonical GEAIRM-owned concepts and record external identities for reused/referenced concepts.

M03-R17 - Identity/label separation: A shared or similar label MUST NOT be used as evidence of semantic identity or equivalence.


44. Normative Force and Model Status​

GEAIRM uses the following normative keywords:

  • MUST / MUST NOT - mandatory requirement;
  • SHOULD / SHOULD NOT - recommended requirement unless a justified exception applies;
  • MAY - permitted option.

The normative status of a model statement is separate from the semantic authority of the concepts it uses.

For example, GEAIRM may normatively require preservation of an AIPS distinction even though AIPS, not GEAIRM, owns the underlying concept definitions.

During the normative programme, model elements may have specification status such as:

  • canonical - accepted by the current GEAIRM normative baseline;
  • referenced - externally authoritative and intentionally used;
  • candidate - unresolved and not yet canonical;
  • deprecated - retained for compatibility but no longer preferred, where later slices establish deprecation;
  • informative - explanatory only.

These are specification statuses, not lifecycle States of the represented AI domain thing.

M03-R18 - Normative/informative separation: Informative examples, diagrams and implementation illustrations MUST NOT override normative definitions, relationships, rules or external semantic authorities.


45. Canonical Element Record Template​

Later GEAIRM slices SHOULD document each canonical or reused L1 model element using a consistent record structure.

A human-readable element record SHOULD contain, where applicable:

FieldPurpose
Preferred labelHuman-readable canonical label.
Element kindConcept, Classification, Role, State, Relationship Type or other M03 metamodel kind.
DefinitionGEAIRM definition for an OWNED element, or authoritative definition/reference for an external element.
Semantic authorityGEAIRM, AIPS, BPS, Agent Ontology, GEDRM, GEARM or another named authority.
Authority identifierStable external identifier/IRI where available.
Representation dispositionOWNED, REUSED, REFERENCED, MAPPED, CANDIDATE or RESERVED.
Normative statuscanonical, referenced, candidate, deprecated or informative.
Source sliceSlice establishing or reconciling the element.
RelationshipsCanonical relationships to other elements.
AliasesHuman-readable aliases that do not create alternate identity.
Constraints / invariantsApplicable semantic conditions or hard distinctions.
NotesExplanatory material that does not change normative meaning.

The later A1 Concept Registry MAY project these fields into machine-readable form after M17 stabilization.


46. Canonical Relationship Record Template​

Each Relationship Type introduced in later slices SHOULD record:

FieldPurpose
Preferred labelCanonical relationship label.
DefinitionExact semantic meaning.
Semantic authorityGEAIRM or external authority.
Subject kindsPermitted subject metamodel/domain kinds.
Object kindsPermitted object metamodel/domain kinds.
DirectionalityDirected or explicitly symmetric.
TransitivityExplicitly true only when authorized; otherwise false/not asserted.
InverseExplicit inverse if one is normatively established.
Non-implicationsCommon invalid conclusions that MUST NOT be drawn.
Source sliceSlice establishing the relationship.

This template is intended to reduce accidental semantic strengthening during later RDF/OWL or SHACL formalization.


47. Illustrative Metamodel Usage​

The following examples demonstrate M03 modelling discipline without defining new substantive AI-domain semantics.

47.1 Reused external Concept​

Element kind: Concept
Preferred label: AI Primitive
Semantic authority: AIPS
Authority identity: existing AIPS identifier
Disposition: REUSED
GEAIRM definition: none - authoritative meaning remains AIPS-owned

47.2 Reused Agent concept​

Element kind: Concept
Preferred label: Agent
Semantic authority: KivaNura Agent Ontology
Disposition: REUSED
GEAIRM definition: none - Agent identity remains externally authoritative

47.3 Unresolved semantic gap​

Working label: Grounding
Disposition: CANDIDATE
Canonical GEAIRM concept: no
Resolution slice: M07

47.4 Architecture-reserved concern​

Working label: Agent Execution Capability
Disposition: RESERVED
Reason: architectural capability, not reference-model semantic concept
Owning future specification: GEAIRA

47.5 Context-qualified assertion​

Subject S
-- Relationship R --> Object O
[within Modelling Context C]

Meaning:
the assertion applies within C;
it does not assert universal applicability outside C.

48. Modelling Anti-Patterns​

The following are explicitly invalid GEAIRM modelling practices.

48.1 Shadow concept creation​

Creating geairm:Agent solely because GEAIRM needs to refer to the Agent Ontology Agent.

48.2 Label-based equivalence​

Declaring two terms equivalent because both contain words such as autonomy, control, model, context or decision.

48.3 Role-as-type modelling​

Turning a contextual Role into an intrinsic class without semantic authority.

48.4 State-as-type modelling​

Creating new intrinsic kinds from lifecycle or currentness States merely for implementation convenience.

48.5 Classification-as-inheritance modelling​

Treating a risk tier, execution mode or other controlled Classification as taxonomic inheritance when its authority does not define that semantics.

48.6 Dependency inheritance​

Assuming that a thing inherits AI Product status, intelligence type, risk, authority or Agent identity from something it depends on.

48.7 Architecture leakage​

Embedding platform plane, ABB, registry or deployment responsibilities inside a GEAIRM concept definition.

48.8 Unqualified universal assertion​

Taking a relationship established only within a particular organizational, temporal or authority scope and treating it as globally true.


49. M03 Canonical Metamodel Summary​

The GEAIRM modelling structure can be summarized as:

                    Reference Model Element
|
+----------+---------+---------+-----------+
| | | |
Concept Classification Role State
|
+------------------+---------------------------+
|
Semantic Assertion
|
Relationship Type
|
optional Modelling Context
|
authority / provenance
|
+------------------+------------------+
| |
External Semantic Reference Evidence Reference
|
Semantic Mapping

Invariant constrains permitted interpretation across the model.

This is a modelling grammar, not an enterprise AI taxonomy. Substantive AI Concepts begin in M04.


50. M03 Normative Rules​

For traceability, M03 establishes the following rules.

RuleRequirement
M03-R01GEAIRM metamodel constructs MUST remain distinct from enterprise AI domain concepts.
M03-R02GEAIRM MUST preserve L0 metamodel, L1 reference-model and L2 scenario/implementation level distinctions.
M03-R03A frequently used AI term MUST NOT become a canonical GEAIRM Concept without satisfying the Concept criterion and M01 semantic-gap discipline.
M03-R04Classification MUST NOT by itself change Concept identity or semantic authority.
M03-R05A Role MUST NOT be treated as an intrinsic type unless the applicable semantic authority explicitly defines that relationship.
M03-R06A State transition MUST NOT be modelled as taxonomic specialization unless explicitly authorized.
M03-R07L0 Modelling Context MUST NOT pre-empt the unresolved L1 enterprise AI Context concept.
M03-R08Relationship inference behavior MUST be explicit; symmetry, transitivity, inverse, equivalence, containment and ownership transfer MUST NOT be assumed.
M03-R09Normative Semantic Assertions SHOULD be traceable to the slice or external authority that establishes them.
M03-R10Externally owned concepts MUST retain authoritative identity where available; GEAIRM MUST NOT create local substitutes merely for convenience.
M03-R11Semantic equivalence MUST NOT be inferred from lexical similarity or overlapping use cases.
M03-R12L0 Evidence Reference MUST NOT be treated as the future L1 AI Evidence domain concept.
M03-R13Representation disposition MUST NOT imply transfer of semantic authority.
M03-R14Taxonomy, Classification, Role and State MUST remain semantically distinct unless explicitly related by an authoritative specification.
M03-R15Context-qualified assertions MUST NOT be generalized beyond their declared scope without semantic justification.
M03-R16Only explicitly authorized inference behavior may be relied upon for GEAIRM interpretation or conformance.
M03-R17Labels and aliases MUST NOT be used as evidence of semantic identity or equivalence.
M03-R18Informative examples, diagrams and implementation illustrations MUST NOT override normative semantics.

51. M03 Exit Criteria​

GEAIRM-M03 is complete when:

  • the L0/L1/L2 modelling levels are explicit;
  • the core reference-model metamodel constructs are defined;
  • Concept, Classification, Role and State are distinguished;
  • Modelling Context is separated from the future L1 Context concept;
  • Relationship Type and Semantic Assertion rules are established;
  • external semantic references and semantic mappings are defined;
  • Evidence Reference is separated from the future L1 AI Evidence concept;
  • representation disposition is separated from semantic authority;
  • inference behavior is conservative by default;
  • identity is separated from human-readable labels;
  • normative and informative content are distinguished;
  • canonical element and relationship record templates are established;
  • architecture leakage and other modelling anti-patterns are explicitly prohibited;
  • all M03 constructs preserve the M01 authority boundary and M02 abstraction boundary.

M03 conclusion: these criteria are satisfied by this slice. GEAIRM now has a stable modelling grammar within which M04 can establish or reference the Core AI & Intelligence Semantics without conflating domain concepts, contextual facets, external authorities or architecture constructs.


52. Forward Work​

The next slice is:

GEAIRM-M04 - Core AI & Intelligence Semantics

M04 will establish or reconcile the fundamental semantic chain around Artificial Intelligence, Intelligence, Intelligence Capability and AI Primitive. It will apply the M03 record and relationship discipline while preserving AIPS authority for existing AI Primitive and Intelligence Capability semantics and applying the M01 semantic-gap test to any new GEAIRM-owned concepts.


53. GEAIRM-M04 - Core AI & Intelligence Semantics​

53.1 Purpose of this slice​

M01 established semantic jurisdiction, M02 established the purpose and scope of GEAIRM, and M03 established the reference-model metamodel and modelling rules. M04 is the first slice to establish substantive enterprise-AI semantics.

M04 resolves the core semantic frame around four terms that recur throughout AIPS and the planned GEAIRM model:

  • Artificial Intelligence;
  • Intelligence;
  • Intelligence Capability;
  • Artificial Intelligence Primitive (AI Primitive).

The purpose of M04 is not to create another AI Product taxonomy. It is to establish the minimum mechanism-neutral semantic nucleus needed by later GEAIRM slices while preserving the AIPS definitions already in force.

M04 therefore answers:

  1. What does GEAIRM mean by Artificial Intelligence at reference-model level?
  2. Does GEAIRM need a standalone ontological entity named Intelligence?
  3. How is the existing AIPS IntelligenceCapability vocabulary used by GEAIRM?
  4. How does an AIPrimitive relate to an Intelligence Capability?
  5. What distinctions must be preserved among AI, capability, primitive, model and Product?

54. Source Semantics and Authority Resolution​

M04 is grounded in the current AIPS baseline established by M01.

The authoritative AIPS sources already establish the following semantics:

  • an AI Product has Artificial Intelligence as its value-bearing essence and a direct or transitive qualifying AI Primitive dependency;
  • an Artificial Intelligence Primitive is a computational capability capable of materially contributing artificial inference, learning, perception, reasoning, generation, planning, optimization, decision formation or intelligent action;
  • AIPS does not restrict AI Primitives to learned models and explicitly includes learned, symbolic, knowledge-based, planning, intelligent-search, optimization and hybrid mechanisms;
  • Intelligence Capability (IC) answers what intelligence is provided and currently includes Perception, Classification, Detection, Prediction, Generation, Recommendation / Ranking, Reasoning, Planning, Optimization and Control;
  • Execution Mode answers how intelligence is invoked or operated;
  • Intelligence Effect answers what the intelligence produces or causes;
  • the presence of an AI Primitive is necessary but not sufficient for AI Product qualification.

M04 preserves those meanings. It does not replace AIPS identifiers or redefine AIPS Product qualification.

The M01 semantic-gap candidates Artificial Intelligence and Intelligence are resolved by this slice as follows:

CandidateM04 resolutionAuthority / disposition
Artificial IntelligenceEstablish a generic GEAIRM reference-model Concept that names the domain phenomenon represented by artificial computational mechanisms materially providing or contributing intelligence capabilities.GEAIRM / OWNED
IntelligenceDo not create a standalone ontological entity in GEAIRM 0.1. Use the term as a collective semantic label for intelligence capabilities and related AI behaviour where prose requires it.GEAIRM editorial term / not independently canonical
Intelligence CapabilityReuse the existing AIPS class and controlled vocabulary.AIPS / REUSED
AI PrimitiveReuse the existing AIPS class and definition.AIPS / REUSED

This resolution is intentionally conservative. GEAIRM does not need a metaphysical or cognitive theory of Intelligence in order to model enterprise AI precisely.

M04-R01 - Authority preservation: M04 MUST preserve AIPS authority for AIPrimitive, IntelligenceCapability and their existing controlled values; M04 MUST NOT mint GEAIRM substitutes for those AIPS concepts.


55. Canonical Core Semantic Frame​

The GEAIRM core semantic frame is:

Artificial Intelligence
|
| is semantically expressed through
v
one or more Intelligence Capabilities
^
|
| may be materially contributed by
|
one or more AI Primitives

This diagram is conceptual rather than taxonomic.

It does not assert that:

  • Artificial Intelligence is a superclass of AI Primitive;
  • Intelligence Capability is a subtype of AI Primitive;
  • every Intelligence Capability requires exactly one AI Primitive;
  • every AI Primitive independently provides a complete Intelligence Capability;
  • the presence of an AI Primitive establishes an AI Product;
  • an AI Product is equivalent to Artificial Intelligence itself.

The semantic nucleus is deliberately many-to-many. A single primitive may contribute to multiple Intelligence Capabilities and one Intelligence Capability may be constituted or supported by multiple primitives and other dependencies.


56. Artificial Intelligence​

56.1 Canonical GEAIRM definition​

Artificial Intelligence is the enterprise-AI domain phenomenon in which artificial computational mechanisms materially provide or contribute one or more Intelligence Capabilities, including capabilities such as perception, classification, detection, prediction, generation, recommendation, reasoning, planning, optimization or control.

This definition is mechanism-neutral and is deliberately aligned to the existing AIPS AI Primitive and Intelligence Capability semantics.

Artificial Intelligence in GEAIRM is a Concept, not a Product classification, architecture capability, implementation technology, vendor category or deployment topology.

56.2 Element record​

FieldValue
Preferred labelArtificial Intelligence
Element kindConcept
DefinitionThe enterprise-AI domain phenomenon in which artificial computational mechanisms materially provide or contribute one or more Intelligence Capabilities.
Semantic authorityGEAIRM
Authority identifierGEAIRM identifier to be minted during executable formalization after M17.
Representation dispositionOWNED
Normative statuscanonical
Source sliceM04
Primary relationshipscharacterized through Intelligence Capability; materially supported/contributed by AI Primitive semantics.
Constraints / invariantsArtificial Intelligence is not synonymous with AI Product, AI Primitive, AI Model, Agent, automation, analytics or a particular technology.

56.3 What this definition does not imply​

Artificial Intelligence does not imply any of the following merely by being present:

  • Product status;
  • AI Product Family qualification;
  • Agent identity;
  • learned-model dependency;
  • autonomy;
  • decision authority;
  • physical embodiment;
  • a specific execution mode;
  • a specific Intelligence Effect;
  • a particular risk tier;
  • a particular architecture or deployment pattern.

Those concerns are orthogonal and are governed by their applicable semantic authorities and later GEAIRM slices.

56.4 Artificial Intelligence and automation​

Automation and Artificial Intelligence overlap in real systems but are not synonyms.

A deterministic workflow, static routing rule or ordinary software automation does not become Artificial Intelligence merely because it executes without human intervention. Within GEAIRM, treatment as Artificial Intelligence requires a material relationship to the AI semantic nucleus established by M04, including an applicable Intelligence Capability and qualifying artificial-intelligence mechanism semantics consistent with AIPS AI Primitive authority.

This rule does not attempt to classify every possible computational technique. It prevents automated from being used as an alias for AI.

M04-R02 - AI/automation distinction: Automation, software execution or absence of human intervention MUST NOT by itself establish Artificial Intelligence.


57. Intelligence as an Editorial Umbrella, Not a Standalone Entity​

The term Intelligence is used extensively in enterprise AI discourse and in AIPS compound terms such as IntelligenceCapability, IntelligenceEffect and IntelligenceBehaviorEvolutionContract.

M04 deliberately does not establish a standalone canonical GEAIRM entity named Intelligence.

The reasons are:

  1. AIPS already provides the operationally useful IntelligenceCapability and IntelligenceEffect semantics.
  2. A standalone Intelligence entity would invite unnecessary metaphysical, cognitive or human-intelligence interpretations outside GEAIRM scope.
  3. GEAIRM needs to model what capability is provided, through what AI mechanisms, and with what effects—not to define a universal theory of intelligence.
  4. Avoiding a standalone entity reduces accidental overlap with GEARM Information/Analytics semantics and the Agent Ontology.

GEAIRM MAY use lowercase intelligence in explanatory prose as a collective label for the capabilities and behaviour associated with Artificial Intelligence, provided the prose does not imply an additional canonical entity.

Examples of acceptable prose include:

  • "the intelligence provided by the Product";
  • "the origin of productized intelligence";
  • "intelligence capabilities";
  • "intelligence effects".

These phrases do not create a geairm:Intelligence class.

M04-R03 - No standalone Intelligence entity: GEAIRM 0.1 MUST NOT infer or mint a standalone canonical Intelligence entity solely from ordinary-language use of the word intelligence.


58. Intelligence Capability​

58.1 Authority and role in GEAIRM​

AIPS already defines IntelligenceCapability as the dimension that answers what intelligence is provided. M04 therefore reuses that authority.

The current canonical AIPS values are:

  • Perception;
  • Classification;
  • Detection;
  • Prediction;
  • Generation;
  • Recommendation / Ranking;
  • Reasoning;
  • Planning;
  • Optimization;
  • Control.

GEAIRM treats this as a reusable Classification dimension for characterizing the kind or kinds of intelligence provided in an enterprise AI context.

58.2 Element record​

FieldValue
Preferred labelIntelligence Capability
Element kindClassification
DefinitionAIPS-controlled dimension answering what intelligence is provided.
Semantic authorityAIPS
Authority identifieraips:IntelligenceCapability
Representation dispositionREUSED
Normative statusreferenced/canonical through AIPS
Source sliceAIPS; integrated by GEAIRM M04
Canonical valuesPerception, Classification, Detection, Prediction, Generation, Recommendation / Ranking, Reasoning, Planning, Optimization, Control.
Constraints / invariantsIntelligence Capability is not Product Type, AI Primitive Type, Execution Mode, Intelligence Effect, Agency/Autonomy or Embodiment.

58.3 Multi-valued capability​

The AIPS baseline explicitly permits multiple Intelligence Capabilities to characterize the same AI Product. GEAIRM generalizes the reference-model interpretation without broadening the AIPS Product property itself: enterprise AI may involve multiple simultaneous capability classifications.

For example:

Perception + Planning + Control

or:

Generation + Reasoning

represent combinations of capability classifications. They do not create new intrinsic Product Types or new AI Primitive Types.

58.4 Capability is not mechanism​

Reasoning, for example, describes the kind of intelligence provided. It does not by itself state whether that reasoning is produced by:

  • a learned model;
  • a symbolic reasoner;
  • a knowledge reasoner;
  • a hybrid mechanism;
  • an Agent composed from several mechanisms.

Likewise, Planning as an Intelligence Capability is not identical to Planner as an AI Primitive Type.

The identical or related English words describe different semantic dimensions.

M04-R04 - Capability/mechanism orthogonality: An Intelligence Capability MUST NOT be treated as an AI Primitive Type merely because the labels are similar or because a primitive commonly contributes to that capability.


59. Artificial Intelligence Primitive​

59.1 AIPS authoritative definition​

GEAIRM reuses the AIPS concept Artificial Intelligence Primitive (AIPrimitive).

The AIPS definition is:

A computational capability capable of materially contributing artificial inference, learning, perception, reasoning, generation, planning, optimization, decision formation or intelligent action.

GEAIRM does not replace or narrow this definition.

59.2 Element record​

FieldValue
Preferred labelArtificial Intelligence Primitive
AliasAI Primitive
Element kindConcept
DefinitionAIPS authoritative definition applies.
Semantic authorityAIPS
Authority identifieraips:AIPrimitive
Representation dispositionREUSED
Normative statusreferenced/canonical through AIPS
Source sliceAIPS; integrated by GEAIRM M04
Constraints / invariantsAI Primitive is not synonymous with AI Model, AI Product, Intelligence Capability or Agent.

59.3 Primitive is broader than learned model​

The AIPS baseline establishes AIModel as both an AI Asset and an AI Primitive. It also establishes non-model primitive types such as symbolic reasoners, knowledge reasoners, planners, intelligent search, optimizers and hybrid intelligence engines.

Therefore:

AI Model  -> may be an AI Primitive

but

AI Primitive != AI Model

More precisely, under the current AIPS ontology AIModel is a subclass of AIPrimitive, while the inverse is not asserted.

59.4 Model Product is not automatically AI Product​

AIPS also establishes ModelProduct as both a BPS Product and an AI Primitive while explicitly stating that a Model Product is not automatically an AI Product.

This yields an important GEAIRM distinction:

Productization of an AI Primitive
does not itself imply
AI Product Family qualification.

AI Product qualification remains governed by the AIPS AQ1-AQ5 rules and version-specific qualification evidence.

59.5 Primitive contribution does not imply exclusive realization​

An AI Primitive may contribute only part of an Intelligence Capability. Multiple primitives may jointly contribute to a capability, and one primitive may contribute to multiple capabilities.

GEAIRM therefore does not impose a one-primitive-to-one-capability cardinality.

M04-R05 - Primitive breadth: GEAIRM MUST preserve the AIPS mechanism-neutral AI Primitive concept and MUST NOT narrow AI Primitive to learned or machine-learning models.

M04-R06 - Primitive/Product distinction: AI Primitive status, including productization as a Model Product, MUST NOT by itself establish AI Product Family qualification.


60. Capability-to-Primitive Relationship​

M04 introduces one GEAIRM-owned semantic relationship required to connect the two AIPS-owned core concepts without changing their authority.

60.1 contributesToIntelligenceCapability​

FieldValue
Preferred labelcontributes to Intelligence Capability
DefinitionRelates an AI Primitive to an Intelligence Capability to which that primitive materially contributes in a declared semantic context.
Semantic authorityGEAIRM
Subject kindsaips:AIPrimitive
Object kindsaips:IntelligenceCapability values
Directionalitydirected: AI Primitive -> Intelligence Capability
Transitivitynot asserted
Symmetryfalse / not asserted
Inversenone normatively established by M04
Source sliceM04

The relationship means that the primitive contributes materially to enabling, performing or constituting the stated capability in the applicable context.

It does not imply that:

  • the primitive exclusively realizes the capability;
  • the primitive independently provides a complete end-user proposition;
  • the capability is intrinsic to every use of that primitive;
  • the primitive is a Product;
  • the primitive is an AI Product;
  • the surrounding system is an Agent;
  • the relationship is transitive through arbitrary dependency chains.

60.2 Why GEAIRM needs this relationship​

The current AIPS property hasIntelligenceCapability is explicitly scoped to AIProductVersion. GEAIRM must also be able to state, at reference-model level, how a non-Product AI Primitive relates to the kind of intelligence it materially contributes without weakening or widening the AIPS Product property.

M04 therefore introduces a separate semantic relationship rather than changing the domain of aips:hasIntelligenceCapability.

M04-R07 - Product-property preservation: GEAIRM MUST NOT broaden the domain of the AIPS hasIntelligenceCapability Product property merely to express non-Product capability relationships.

M04-R08 - Non-exclusive contribution: contributesToIntelligenceCapability MUST NOT be interpreted as exclusive, exhaustive or one-to-one realization of the capability.


61. Capability, Primitive, Execution and Effect Are Orthogonal Dimensions​

M04 establishes the core distinction among four questions already implicit in AIPS:

DimensionQuestionExample
Intelligence CapabilityWhat kind of intelligence is provided?Reasoning, Planning, Generation
AI PrimitiveWhat qualifying computational mechanism materially contributes AI?Learned Model, Symbolic Reasoner, Planner
Execution ModeHow is the intelligence invoked or operated?Batch, Interactive, Continuous, Agentic / Goal-driven
Intelligence EffectWhat does the intelligence produce or cause?Score, Recommendation, Plan, Digital Action

M04 defines only the first two in detail. Execution Mode and Intelligence Effect are retained as AIPS-owned concepts and are developed further in M06.

The dimensions are not interchangeable.

For example:

Planner
primitive/mechanism dimension

Planning
Intelligence Capability dimension

Plan
Intelligence Effect dimension

These three terms are intentionally distinct even though they share a linguistic root.

Likewise:

Classification
Intelligence Capability

ClassificationEffect
Intelligence Effect

No equivalence is implied by lexical similarity.

M04-R09 - Orthogonal-dimension discipline: Intelligence Capability, AI Primitive, Execution Mode and Intelligence Effect MUST remain distinct semantic dimensions unless an authoritative specification explicitly establishes a narrower relationship.


62. Inference and Learning Are Not Automatically Intelligence Capability Values​

The AIPS AI Primitive definition includes material contribution to inference and learning, but the current AIPS Intelligence Capability controlled vocabulary does not list Inference or Learning as Intelligence Capability values.

GEAIRM MUST preserve that distinction.

M04 therefore does not add Inference or Learning to the AIPS Intelligence Capability vocabulary.

Later slices may model inference, learning or evolution relationships where required, but they MUST NOT silently extend the AIPS controlled vocabulary without an explicit AIPS change or governed GEAIRM relationship that preserves the external classification authority.

M04-R10 - Controlled-vocabulary integrity: The presence of a term in the AI Primitive definition MUST NOT be treated as evidence that it is an AIPS Intelligence Capability value.


63. Artificial Intelligence Does Not Imply Productization​

GEAIRM describes enterprise AI more broadly than qualified AI Products.

The following may participate in an enterprise AI semantic graph without themselves being AI Products:

  • an AI Primitive;
  • an AI Model;
  • an AI Asset;
  • a Model Product;
  • an Agent;
  • a Tool;
  • Data or a Data Product;
  • knowledge or rules;
  • an evaluation artifact;
  • an environment;
  • an AI Product.

This preserves the distinction between AI participation and AI Product identity.

AIPS qualification remains the only authority in the current specification family for deciding whether a Product qualifies as an AI Product.

M04-R11 - Participation/Product boundary: Participation in Artificial Intelligence, use of an AI Primitive, or provision of an Intelligence Capability MUST NOT by itself establish Product status or AI Product Family membership.


64. Artificial Intelligence Does Not Imply Agenthood​

Artificial Intelligence and Agent are different semantic concerns.

An AI capability may be invoked through a non-Agent interface such as a synchronous API, batch process or embedded component. Conversely, an Agent may depend on one or more AI Primitives, Products, rules or tools.

GEAIRM therefore does not define:

Artificial Intelligence IS-A Agent

or:

Agent IS-A Artificial Intelligence

Agent identity remains governed by the KivaNura Agent Ontology. Relationships among Agent, Intelligence Capability and AI Primitive are deferred to M08.

M04-R12 - AI/Agent distinction: Artificial Intelligence MUST NOT be treated as synonymous with Agent, and AI Primitive status MUST NOT establish Agent identity.


65. Artificial Intelligence and Analytics / Information​

GEAIRM does not assert that Artificial Intelligence is equivalent to Analytics, Information or Information Product.

The domains can interact:

  • analytics may invoke AI mechanisms;
  • AI may produce scores, classifications, recommendations or generated content;
  • an Information Product may incorporate AI-derived information;
  • an AI Product may consume Data Products or Information Products.

However:

Artificial Intelligence != Analytics
Artificial Intelligence != Information
AI Product != Information Product

unless an external authority establishes a more specific relationship for a particular entity or context.

GEARM and GEARA remain authoritative for their respective Analytics / Information semantic and architectural concerns.

M04-R13 - Cross-domain non-collapse: GEAIRM MUST NOT collapse Artificial Intelligence into Analytics, Information or Information Product merely because outputs or mechanisms overlap.


66. Core Hard Distinctions Established by M04​

The following distinctions are normative for subsequent GEAIRM slices:

Artificial Intelligence != AI Product
Artificial Intelligence != AI Primitive
Artificial Intelligence != AI Model
Artificial Intelligence != Agent
Artificial Intelligence != Automation

Intelligence Capability != AI Primitive
Intelligence Capability != AI Primitive Type
Intelligence Capability != Execution Mode
Intelligence Capability != Intelligence Effect
Intelligence Capability != AI Product Type

AI Primitive != AI Model
AI Primitive != AI Product
AI Primitive != Agent

AI Model != AI Product
Model Product != AI Product automatically

Planning capability != Planner primitive != Plan effect
Classification capability != Classification Effect

The symbol != denotes semantic non-equivalence, not disjointness in every possible formal sense. M04 does not assert OWL disjointness unless a later executable artifact is explicitly authorized to do so.

This distinction between non-equivalence and formal disjointness is required by M03's conservative inference discipline.


67. Canonical Core Reference View​

The M04 reference view is:

                     +----------------------------+
| Artificial Intelligence |
| GEAIRM-owned |
+-------------+--------------+
|
characterized through
|
v
+----------------------------------+
| Intelligence Capability |
| AIPS-owned |
| |
| Perception | Classification |
| Detection | Prediction |
| Generation | Recommendation |
| Reasoning | Planning |
| Optimization| Control |
+-----------------^----------------+
|
materially contributes to
|
+-----------------+----------------+
| AI Primitive |
| AIPS-owned |
| |
| learned / symbolic / knowledge |
| planning / search / optimization |
| hybrid mechanisms |
+----------------------------------+

The bottom-row primitive labels are illustrative summaries of the AIPS primitive taxonomy. M05 will define the primitive/mechanism model in detail.

A Product boundary may wrap or depend on parts of this graph, but Product identity is deliberately absent from the core view because Product qualification is orthogonal.


68. Illustrative Semantic Examples​

The following examples illustrate M04 without changing AIPS qualification semantics.

68.1 Learned fraud model​

AI Model
IS-A AI Primitive [AIPS]
contributesToIntelligenceCapability Detection
contributesToIntelligenceCapability Prediction

This does not by itself say that the model is an AI Product.

68.2 Symbolic policy reasoner​

Symbolic Reasoner
AI Primitive Type [AIPS]

A particular qualifying primitive
contributesToIntelligenceCapability Reasoning

No learned model is required by this semantic statement.

68.3 Planner​

Planner primitive
contributesToIntelligenceCapability Planning

Planning = capability classification
Planner = primitive type
Plan = Intelligence Effect

The three concepts remain distinct.

68.4 Model Product​

Model Product
IS-A Product [BPS/AIPS]
IS-A AI Primitive [AIPS]

but

Model Product
does not automatically qualify as
AI Product

68.5 Automated workflow without qualifying AI semantics​

Deterministic workflow
executes automatically
routes known cases
contains no qualifying AI Primitive relationship

=> automation alone does not establish Artificial Intelligence

This example does not prohibit deterministic or symbolic AI. A symbolic reasoner can qualify as an AI Primitive under AIPS; the distinction is between ordinary automation and qualifying AI mechanism semantics, not between deterministic and probabilistic computation.


69. M04 Canonical Element Inventory​

M04 establishes or integrates the following core elements:

ElementM03 kindAuthorityDispositionM04 status
Artificial IntelligenceConceptGEAIRMOWNEDcanonical
Intelligenceeditorial umbrella termGEAIRM editorial usagenot a canonical model elementresolved - no standalone entity
Intelligence CapabilityClassificationAIPSREUSEDcanonical external
PerceptionClassification valueAIPSREUSEDcanonical external
ClassificationClassification valueAIPSREUSEDcanonical external
DetectionClassification valueAIPSREUSEDcanonical external
PredictionClassification valueAIPSREUSEDcanonical external
GenerationClassification valueAIPSREUSEDcanonical external
Recommendation / RankingClassification valueAIPSREUSEDcanonical external
ReasoningClassification valueAIPSREUSEDcanonical external
PlanningClassification valueAIPSREUSEDcanonical external
OptimizationClassification valueAIPSREUSEDcanonical external
ControlClassification valueAIPSREUSEDcanonical external
Artificial Intelligence PrimitiveConceptAIPSREUSEDcanonical external
contributesToIntelligenceCapabilityRelationship TypeGEAIRMOWNEDcanonical

Supporting AIPS concepts such as AIModel, ModelProduct, ExecutionMode, IntelligenceEffect and AIPrimitiveType are referenced by M04 for boundary clarity but are not exhaustively developed until their designated slices.


70. M04 Normative Rules​

For traceability, M04 establishes the following rules.

RuleRequirement
M04-R01M04 MUST preserve AIPS authority for AIPrimitive, IntelligenceCapability and their existing controlled values.
M04-R02Automation, software execution or absence of human intervention MUST NOT by itself establish Artificial Intelligence.
M04-R03GEAIRM 0.1 MUST NOT mint a standalone canonical Intelligence entity solely from ordinary-language use of the term.
M04-R04Intelligence Capability MUST NOT be treated as AI Primitive Type merely because labels or common implementations overlap.
M04-R05GEAIRM MUST preserve the mechanism-neutral AIPS AI Primitive concept and MUST NOT narrow it to learned/ML models.
M04-R06AI Primitive status or Productization of a primitive MUST NOT by itself establish AI Product Family qualification.
M04-R07GEAIRM MUST NOT broaden the domain of AIPS hasIntelligenceCapability merely to express non-Product relationships.
M04-R08contributesToIntelligenceCapability MUST NOT be interpreted as exclusive, exhaustive, transitive or one-to-one realization.
M04-R09Intelligence Capability, AI Primitive, Execution Mode and Intelligence Effect MUST remain distinct semantic dimensions.
M04-R10Terms appearing in the AI Primitive definition MUST NOT automatically be added to the AIPS Intelligence Capability controlled vocabulary.
M04-R11Participation in AI, use of an AI Primitive or provision of an Intelligence Capability MUST NOT by itself establish Product or AI Product status.
M04-R12Artificial Intelligence MUST NOT be treated as synonymous with Agent, and AI Primitive status MUST NOT establish Agent identity.
M04-R13GEAIRM MUST NOT collapse Artificial Intelligence into Analytics, Information or Information Product.
M04-R14Artificial Intelligence, AI Primitive, AI Model, AI Product and Intelligence Capability MUST remain non-equivalent unless an authoritative later specification establishes a narrower mapping.
M04-R15M04 semantic non-equivalence MUST NOT be interpreted as formal OWL disjointness unless explicitly authorized during executable formalization.
M04-R16M04 MUST preserve the distinction Planning capability != Planner primitive != Plan effect.
M04-R17M04 MUST preserve the distinction between Classification Intelligence Capability and ClassificationEffect Intelligence Effect.
M04-R18Product qualification semantics remain authoritative in AIPS and MUST NOT be inferred from the M04 core semantic graph.

71. M04 Exit Criteria​

GEAIRM-M04 is complete when:

  • the M01 Artificial Intelligence semantic-gap candidate is resolved;
  • the M01 Intelligence semantic-gap candidate is resolved without unnecessary ontology expansion;
  • Artificial Intelligence has a canonical mechanism-neutral GEAIRM definition;
  • AIPS Intelligence Capability is explicitly reused rather than redefined;
  • the current Intelligence Capability controlled values are preserved;
  • AIPS AI Primitive is explicitly reused rather than redefined;
  • the mechanism-neutral Primitive boundary is preserved;
  • Intelligence Capability and AI Primitive are semantically separated;
  • the Product-scoped AIPS hasIntelligenceCapability property is not broadened;
  • a generic primitive-to-capability relationship is defined without changing AIPS ownership;
  • AI, Product, Agent, Model, capability, primitive, execution and effect boundaries are explicit;
  • inference and learning are not silently added to the Intelligence Capability vocabulary;
  • automation is not conflated with AI;
  • AI is not collapsed into Analytics or Information;
  • M03 conservative inference rules remain intact;
  • M04 does not pre-empt the detailed primitive taxonomy planned for M05 or Execution/Effect semantics planned for M06.

M04 conclusion: these criteria are satisfied by this slice. GEAIRM now has a stable core semantic nucleus on which M05 can build the detailed AI Primitive & Intelligence Mechanism Model.


72. Forward Work​

The next slice is:

GEAIRM-M05 - AI Primitive & Intelligence Mechanism Model

M05 will elaborate the AIPS AI Primitive taxonomy and relationships among learned models, symbolic reasoners, knowledge reasoners, planners, intelligent search, optimizers and hybrid intelligence mechanisms. It will preserve the M04 distinction between the Intelligence Capability being provided and the mechanism that materially contributes it, and it will not convert primitive composition into Product qualification.


73. GEAIRM-M05 - AI Primitive & Intelligence Mechanism Model​

73.1 Purpose of this slice​

M04 established the core semantic nucleus around Artificial Intelligence, Intelligence Capability and AI Primitive. M05 now elaborates the mechanism side of that nucleus without changing the AIPS-owned AI Primitive semantics.

M05 answers five questions:

  1. How does GEAIRM represent the existing AIPS AI Primitive Type classification?
  2. What distinctions must be preserved among Learned Model, Symbolic Reasoner, Knowledge Reasoner, Planner, Intelligent Search, Optimizer and Hybrid Intelligence Engine?
  3. How does an AI Primitive Type differ from an AI Primitive, AI Model, Intelligence Capability, Intelligence Effect or Product classification?
  4. How may a primitive be related to its primitive-type classification without broadening the AIPS Product-qualification evidence model?
  5. What mechanism-composition statements may M05 make without pre-empting the dependency and composition semantics reserved for M11?

M05 remains mechanism-neutral. It does not rank learned mechanisms above symbolic, knowledge-based, planning, search, optimization or hybrid mechanisms, and it does not use presence of a trained model as the defining test for Artificial Intelligence.


74. Source Semantics and Authority Resolution​

AIPS already establishes the authoritative AI Primitive concept and current AI Primitive Type vocabulary.

The current AIPS AI Product ontology identifies the following typical primitive classes in explanatory prose:

  • Learned Model;
  • Symbolic Reasoner;
  • Knowledge Reasoner;
  • Planner;
  • Intelligent Search;
  • Optimizer;
  • Hybrid Intelligence Engine.

The AIPS machine-readable core represents these as controlled values of aips:AIPrimitiveType:

aips:LEARNED_MODEL
aips:SYMBOLIC_REASONER
aips:KNOWLEDGE_REASONER
aips:PLANNER
aips:INTELLIGENT_SEARCH
aips:OPTIMIZER
aips:HYBRID_INTELLIGENCE_ENGINE

M05 preserves this authority model. It does not mint GEAIRM replacements for those values.

The phrase intelligence mechanism is used in M05 as an explanatory umbrella for the computational mechanism represented by one or more AI Primitives. M05 does not establish a new canonical GEAIRM class named IntelligenceMechanism, because the existing AIPS AIPrimitive concept already provides the required reference-model anchor.

M05-R01 - Primitive authority preservation: GEAIRM MUST preserve AIPS authority for AIPrimitive, AIPrimitiveType and the current AIPS AI Primitive Type values.

M05-R02 - No redundant mechanism class: GEAIRM 0.1 MUST NOT mint a standalone IntelligenceMechanism concept merely as an alias or superclass for AIPS AIPrimitive.


75. AI Primitive vs AI Primitive Type​

M05 makes explicit a distinction that is essential to the reference model:

AI Primitive
= the qualifying computational capability / mechanism

AI Primitive Type
= the controlled classification applied to that primitive or to evidence about it

An AI Primitive Type is therefore not itself the executing primitive.

For example:

FraudModel-4.1
may be an AI Primitive

LEARNED_MODEL
is an AI Primitive Type classification value

Likewise:

PolicyReasoner-X
may be an AI Primitive

SYMBOLIC_REASONER
is an AI Primitive Type classification value

The current AIPS RDF models aips:AIPrimitiveType as a class whose controlled values are individuals. M05 therefore treats AI Primitive Type as an M03 Classification, not as a taxonomic superclass hierarchy for primitive resources.

M05-R03 - Primitive/type separation: AIPrimitive and AIPrimitiveType MUST remain distinct; an AI Primitive Type value MUST NOT be treated as the primitive resource itself.

M05-R04 - Classification not inheritance: Classification of an AI Primitive by an AI Primitive Type MUST NOT be modelled as subclass inheritance unless AIPS explicitly establishes such an axiom.


76. AI Primitive Type​

76.1 Element record​

FieldValue
Preferred labelAI Primitive Type
Element kindClassification
DefinitionAIPS-controlled classification identifying the qualifying mechanism type recorded for AI Primitive evidence and reused by GEAIRM for mechanism characterization.
Semantic authorityAIPS
Authority identifieraips:AIPrimitiveType
Representation dispositionREUSED
Normative statusreferenced/canonical through AIPS
Source sliceAIPS; integrated by GEAIRM M05
Canonical valuesLEARNED_MODEL, SYMBOLIC_REASONER, KNOWLEDGE_REASONER, PLANNER, INTELLIGENT_SEARCH, OPTIMIZER, HYBRID_INTELLIGENCE_ENGINE
Constraints / invariantsAI Primitive Type is not Intelligence Capability, Intelligence Effect, Execution Mode, AI Product Type, Agent type or Product Family.

76.2 Controlled vocabulary boundary​

M05 treats the seven values above as the current AIPS controlled vocabulary. It does not declare them philosophically exhaustive of every possible future AI mechanism and does not add new primitive types.

A new primitive type requires an AIPS change or an explicitly governed future extension. GEAIRM MUST NOT extend the AIPS list merely because an implementation uses a new vendor term or algorithm family.

76.3 Evidence cardinality is not universal mechanism cardinality​

The current AIPS SHACL model requires one primitiveType value on an AIPrimitiveEvidence record. That rule applies to the AIPS qualification-evidence structure.

GEAIRM MUST NOT infer from that evidence-cardinality rule that every AI Primitive can have exactly one mechanism classification in every possible semantic context. M05 introduces no universal functional-cardinality axiom for primitive classification.

M05-R05 - Vocabulary integrity: GEAIRM MUST NOT silently add values to the AIPS AI Primitive Type vocabulary.

M05-R06 - Evidence/cardinality boundary: Cardinality constraints on AIPS AIPrimitiveEvidence.primitiveType MUST NOT be generalized into universal cardinality constraints on AI Primitive semantics unless AIPS explicitly establishes that broader rule.


77. Primitive-Type Classification Relationship​

The AIPS property primitiveType is scoped to AIPrimitiveEvidence. GEAIRM also needs to characterize an AI Primitive directly at reference-model level without changing the AIPS evidence property.

M05 therefore introduces a separate GEAIRM-owned relationship.

77.1 classifiedAsPrimitiveType​

FieldValue
Preferred labelclassified as AI Primitive Type
DefinitionRelates an AI Primitive to an AIPS AI Primitive Type value that characterizes the primitive's qualifying mechanism in a declared semantic context.
Semantic authorityGEAIRM
Subject kindsaips:AIPrimitive
Object kindsvalues of aips:AIPrimitiveType
Directionalitydirected: AI Primitive -> AI Primitive Type
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted by M05
Inversenone normatively established by M05
Source sliceM05

The relationship does not replace aips:primitiveType. The two serve different semantic scopes:

aips:primitiveType
AIPrimitiveEvidence -> AIPrimitiveType
qualification-evidence scope

geairm:classifiedAsPrimitiveType
AIPrimitive -> AIPrimitiveType
generic reference-model scope

The future executable GEAIRM namespace and identifier are not minted until the A1-A8 formalization programme after M17.

M05-R07 - Evidence-property preservation: GEAIRM MUST NOT broaden the domain of AIPS primitiveType from AIPrimitiveEvidence to AIPrimitive merely for convenience.

M05-R08 - Classification relationship restraint: classifiedAsPrimitiveType MUST NOT imply Product status, AI Product qualification, Intelligence Capability, Agent identity, risk tier or architecture realization.


78. Learned Model​

LEARNED_MODEL is an AIPS AI Primitive Type value. The AIPS ontology gives examples including LLMs, classifiers, vision models and embedding models.

M05 uses the classification to characterize an AI Primitive whose qualifying mechanism is materially embodied in a learned model capability. This is a GEAIRM reference-model interpretation of the AIPS value, not a replacement definition of the AIPS term.

Important distinctions are:

LEARNED_MODEL
= AI Primitive Type classification

AIModel
= AIPS class that is both AIAsset and AIPrimitive

ModelProduct
= BPS Product + AIPS AIPrimitive
= not automatically an AI Product

The current AIPS RDF does not establish a universal ontology axiom stating that every aips:AIModel is automatically related to aips:LEARNED_MODEL through a primitive-type property. Where the classification is material to an assertion, it SHOULD be explicit rather than inferred solely from similar terminology.

A learned model may contribute to one or more Intelligence Capabilities. The specific capability is not determined by the LEARNED_MODEL classification alone.

For example, learned models may materially contribute to Detection, Prediction, Generation, Perception, Recommendation or other applicable Intelligence Capability values depending on the actual primitive.

M05-R09 - Learned-model non-collapse: LEARNED_MODEL, AIModel and ModelProduct MUST remain distinct semantic constructs.

M05-R10 - Type does not determine capability: Classification as LEARNED_MODEL MUST NOT by itself determine which Intelligence Capability the primitive contributes.


79. Symbolic Reasoner​

SYMBOLIC_REASONER is an AIPS AI Primitive Type value. AIPS gives an expert-system inference engine as a representative example.

M05 treats the value as characterizing a qualifying AI Primitive whose material mechanism is symbolic reasoning rather than learned-model behaviour.

The classification does not imply that every rule engine, decision table, conditional branch or deterministic program is an AI Primitive.

This distinction is important because AIPS simultaneously permits symbolic reasoners as AI Primitives and provides boundary examples in which ordinary deterministic if/else logic may remove the only qualifying AI Primitive from an AI Product constitution.

Therefore:

Deterministic
does not imply non-AI

but

Deterministic automation
does not imply symbolic AI

The controlling criterion remains the AIPS AI Primitive definition and the material intelligence behaviour of the mechanism, not determinism versus probabilism.

A Symbolic Reasoner classification also does not prescribe which Intelligence Capability applies. A symbolic reasoner may commonly contribute to Reasoning, Classification, Planning, Control or other capabilities, but those capability relationships require their own semantic assertion.

M05-R11 - Symbolic/automation boundary: Ordinary deterministic rules or routing MUST NOT be classified as SYMBOLIC_REASONER solely because they manipulate conditions or symbols.

M05-R12 - Determinism neutrality: Determinism MUST NOT by itself disqualify a mechanism from AI Primitive status where the AIPS AI Primitive semantics are otherwise satisfied.


80. Knowledge Reasoner​

KNOWLEDGE_REASONER is an AIPS AI Primitive Type value. AIPS gives ontology and knowledge-graph reasoning engines as representative examples.

M05 distinguishes the reasoning mechanism from the knowledge resources it may consume.

Knowledge Graph / Ontology / Knowledge Resource
may be input or dependency

Knowledge Reasoner
is the qualifying computational mechanism

A knowledge store, ontology document or graph dataset does not become a Knowledge Reasoner merely because a reasoner operates over it.

Likewise, the Knowledge Reasoner classification does not transfer AI Primitive status to all knowledge resources on which it depends.

The semantic treatment of Knowledge itself is deferred to M07, and detailed dependency semantics are deferred to M11.

M05-R13 - Knowledge/mechanism separation: Knowledge resources and KNOWLEDGE_REASONER primitives MUST remain distinct; dependency on or use by a Knowledge Reasoner MUST NOT transfer AI Primitive status to the knowledge resource.


81. Planner​

PLANNER is an AIPS AI Primitive Type value. AIPS gives a goal/action planner as a representative example.

M04 already established the hard distinction:

Planner
primitive/mechanism dimension

Planning
Intelligence Capability dimension

Plan
Intelligence Effect dimension

M05 strengthens the mechanism side of that distinction.

A primitive classified as PLANNER may materially contribute to the Planning Intelligence Capability, but this mapping is not an identity relation and is not asserted as universally exclusive. A planner may also contribute to other capabilities, and Planning capability may be provided through other mechanism compositions.

The existence of a Plan effect does not prove that the producing primitive is classified as PLANNER.

M05-R14 - Planner/capability/effect separation: PLANNER, Planning and Plan MUST remain distinct and MUST NOT be inferred from one another solely by lexical similarity.


INTELLIGENT_SEARCH is an AIPS AI Primitive Type value. AIPS gives heuristic or state-space search capability as representative examples.

M05 distinguishes intelligent-search mechanism semantics from ordinary retrieval, lookup or indexing.

An enterprise system MAY perform search without that search being an AI Primitive. Classification as INTELLIGENT_SEARCH requires the search capability itself to satisfy the AIPS AI Primitive semantics and materially constitute the qualifying intelligence behaviour.

The M07 semantic-gap candidate Retrieval therefore remains unresolved by M05. Retrieval used for grounding or context assembly is not automatically an Intelligent Search primitive.

Retrieval
!= automatically
INTELLIGENT_SEARCH

The relationship may exist in a particular implementation, but it requires explicit semantic justification.

M05-R15 - Search/retrieval distinction: Ordinary retrieval or lookup MUST NOT be treated as INTELLIGENT_SEARCH solely because a search operation occurs.


83. Optimizer​

OPTIMIZER is an AIPS AI Primitive Type value. The AIPS ontology explicitly qualifies the example as an optimization engine where it constitutes intelligent behaviour.

This qualifier is normative for GEAIRM interpretation.

Optimization as a general mathematical or software technique does not automatically establish AI Primitive status. The optimizer must materially participate in the qualifying artificial-intelligence behaviour under the AIPS AI Primitive definition.

M04 also distinguishes:

OPTIMIZER
AI Primitive Type

Optimization
Intelligence Capability

The two may be related in a scenario but are not identical.

M05-R16 - Optimizer qualification: Use of optimization mathematics or an optimization library MUST NOT by itself establish the OPTIMIZER AI Primitive Type; the optimization mechanism must materially constitute qualifying intelligent behaviour.

M05-R17 - Optimizer/capability separation: OPTIMIZER and Optimization MUST remain distinct primitive-type and capability dimensions.


84. Hybrid Intelligence Engine​

HYBRID_INTELLIGENCE_ENGINE is an AIPS AI Primitive Type value. AIPS describes it through learned plus symbolic/reasoning mechanisms.

M05 interprets the classification as applying where the qualifying primitive itself materially integrates more than one materially distinct intelligence-mechanism style into a coherent intelligence mechanism.

The classification MUST NOT be inferred merely because a wider system contains multiple AI Primitives.

In particular:

Hybrid Intelligence Engine
!= Composite Derived AI Product
!= BPS Compound Product
!= multi-Agent system
!= ensemble merely because several components are present
!= orchestration merely because several Products or tools are invoked

Those are different semantic axes.

Detailed dependency, composition and federation semantics remain reserved for M11. M05 only establishes that hybrid primitive classification is a mechanism classification, not a Product-topology classification.

M05-R18 - Hybrid/composition boundary: HYBRID_INTELLIGENCE_ENGINE MUST NOT be inferred solely from the presence of multiple primitives, Products, Agents or dependencies in a wider system.

M05-R19 - Hybrid/Product boundary: Hybrid Intelligence Engine classification MUST remain distinct from Composite derivation, Compound Product structure and AI Product Type.


85. Primitive Type Does Not Determine Intelligence Capability​

AI Primitive Type and Intelligence Capability are orthogonal dimensions.

A mechanism classification answers:

What kind of qualifying computational mechanism is this?

An Intelligence Capability classification answers:

What kind of intelligence does it materially provide or contribute?

The reference-model relationship is therefore many-to-many in principle:

AI Primitive
|\
| \ classifiedAsPrimitiveType
| \
| v
| AI Primitive Type
|
+---- contributesToIntelligenceCapability ----> Intelligence Capability

Illustrative mappings may include:

Learned Model      -> Detection / Prediction / Generation / ...
Symbolic Reasoner -> Reasoning / Classification / Control / ...
Knowledge Reasoner -> Reasoning / ...
Planner -> Planning / ...
Intelligent Search -> applicable capability depending on use
Optimizer -> Optimization / ...
Hybrid Engine -> multiple applicable capabilities

The ellipses are deliberate. They indicate that M05 does not define a closed mapping matrix.

GEAIRM MUST NOT infer an Intelligence Capability solely from AI Primitive Type. The explicit M04 contributesToIntelligenceCapability relationship remains the canonical mechanism-to-capability assertion.

M05-R20 - No type-to-capability inference: AI Primitive Type MUST NOT deterministically imply a specific Intelligence Capability unless an explicit relationship is asserted or an authoritative future rule establishes the mapping.


86. Primitive Type Does Not Determine Intelligence Effect or Execution Mode​

The mechanism used does not by itself determine either how the intelligence is invoked or what effect it produces.

For example:

Learned Model
may be invoked interactively, continuously, in batch or through an Agentic path

Symbolic Reasoner
may produce a classification, recommendation, decision-related result or another effect

The exact Execution Mode and Intelligence Effect semantics remain AIPS-owned and are developed in M06.

M05 therefore establishes:

AI Primitive Type
!= Execution Mode
!= Intelligence Effect

and no default mapping is created between these dimensions.

M05-R21 - Execution/effect orthogonality: AI Primitive Type MUST NOT by itself determine Execution Mode or Intelligence Effect.


87. Primitive Identity, Asset Identity and Product Identity​

AIPS permits one resource to participate in more than one semantic category.

The current AIPS ontology states:

AIModel
IS-A AIAsset
IS-A AIPrimitive

ModelProduct
IS-A BPS Product
IS-A AIPrimitive

This is not a contradiction. It means the same resource may have both artifact/Product identity and primitive participation where the authoritative ontology establishes those types.

GEAIRM MUST therefore avoid the false assumption that AIPrimitive means "non-asset" or "non-Product".

However, the converse also remains false:

AIAsset      != automatically AIPrimitive
Product != automatically AIPrimitive
AIPrimitive != automatically AIProduct

For example, AIPS AIDataset, AIPipeline and AIEvalReport are AI Assets but are not declared subclasses of AIPrimitive in the current core ontology.

M05-R22 - Multi-typing preservation: GEAIRM MUST preserve authoritative multi-typing such as AIModel being both AIAsset and AIPrimitive and ModelProduct being both Product and AIPrimitive.

M05-R23 - No converse inference: Asset status, Product status or participation in an AI system MUST NOT be used to infer AI Primitive status unless the authoritative semantics establish it.


88. Primitive Composition, Dependency and Hybridization Boundary​

M05 deliberately does not define the full composition graph for AI Primitives.

The current AIPS vocabulary already contains generic dependency relationships such as dependsOn and materiallyDependsOn, while Product composability distinguishes operational composition from creation of a new Derived AI Product.

The detailed generic GEAIRM semantics for:

  • direct dependency;
  • transitive dependency;
  • primitive composition;
  • federation;
  • multi-Agent composition;
  • upstream Product dependency;
  • dependency provenance;
  • topology;

remain assigned to M11.

M05 establishes only three boundary principles:

  1. multiple primitives used together do not automatically become one Hybrid Intelligence Engine;
  2. a Hybrid Intelligence Engine does not automatically create a new Product;
  3. a primitive's dependencies do not transfer their primitive type, Product Family, risk classification or semantic authority to the dependent resource.

M05-R24 - Composition deferral: M05 MUST NOT introduce generic primitive-composition or dependency-transitivity rules that pre-empt M11.


89. Primitive Qualification Is Contextual to the Mechanism, Not the Vendor Label​

An implementation technology, library, framework or vendor Product name does not establish AI Primitive Type by itself.

Examples of invalid inference include:

uses an LLM API
=> therefore the calling component is a Learned Model primitive

uses a knowledge graph
=> therefore the calling component is a Knowledge Reasoner

uses a search engine
=> therefore the calling component is Intelligent Search

uses an optimizer library
=> therefore the calling component is an Optimizer primitive

The primitive classification applies to the resource that actually constitutes the qualifying computational mechanism in the declared semantic context.

This rule is consistent with BPS composition non-inheritance and M03 dependency-inference restraint.

M05-R25 - No dependency type inheritance: A component MUST NOT inherit the AI Primitive Type of a dependency merely because it invokes, contains or depends on that primitive.


90. M05 Canonical Primitive Reference View​

The M05 reference view is:

                             Artificial Intelligence
|
v
Intelligence Capability
AIPS-owned
^
|
contributesToIntelligenceCapability
|
+-------+-------+
| AI Primitive |
| AIPS-owned |
+-------+-------+
|
classifiedAsPrimitiveType
|
v
+-------------------------+
| AI Primitive Type |
| AIPS-owned |
+------------+------------+
|
+-------------+---------------+---------------+-------------+
| | | | |
LEARNED_MODEL SYMBOLIC_REASONER KNOWLEDGE_REASONER PLANNER INTELLIGENT_SEARCH
|
+-----------------+-----------------------+
| |
OPTIMIZER HYBRID_INTELLIGENCE_ENGINE

The layout does not imply subclass relations among the seven controlled values. They are shown as members of the AIPS AI Primitive Type classification.

The reference view also deliberately omits Product boundaries, Agent relationships, Data/Knowledge dependencies and runtime topology; those concerns are orthogonal or assigned to later slices.


91. Illustrative Semantic Examples​

91.1 Learned fraud model​

FraudDetector-v4.1
IS-A aips:AIModel [AIPS]
therefore IS-A aips:AIPrimitive [AIPS]
classifiedAsPrimitiveType LEARNED_MODEL [GEAIRM]
contributesToIntelligenceCapability Detection
contributesToIntelligenceCapability Prediction

The example does not imply that the model is an AI Product.

91.2 Symbolic policy reasoner​

PolicyReasoner-X
IS-A AIPrimitive
classifiedAsPrimitiveType SYMBOLIC_REASONER
contributesToIntelligenceCapability Reasoning

Its deterministic implementation does not invalidate AI Primitive status merely because it is deterministic.

91.3 Knowledge resource plus reasoner​

PolicyKnowledgeGraph
knowledge resource
|
| used by
v
KnowledgeReasoner-X
AIPrimitive
classifiedAsPrimitiveType KNOWLEDGE_REASONER

The knowledge graph does not inherit AI Primitive status from the reasoner.

91.4 Planner producing a plan​

Planner-X
AIPrimitive
classifiedAsPrimitiveType PLANNER
contributesToIntelligenceCapability Planning
|
v
may produce Plan effect

Planner, Planning and Plan remain separate dimensions.

91.5 Search used for grounding​

Retriever-X
performs retrieval

No conclusion:
INTELLIGENT_SEARCH

If the retrieval mechanism itself satisfies AIPS AI Primitive semantics and is appropriately classified, an explicit primitive-type assertion may be added. M05 does not infer it from the word retrieval.

91.6 Hybrid mechanism vs composed system​

System S
depends on LearnedModel-A
depends on SymbolicReasoner-B

No automatic conclusion:

System S IS-A Hybrid Intelligence Engine
System S IS-A AI Product

Those classifications require their own authoritative semantics.


92. M05 Canonical Element Inventory​

M05 establishes or integrates the following model elements:

ElementM03 kindAuthorityDispositionM05 status
AI PrimitiveConceptAIPSREUSEDcanonical external
AI Primitive TypeClassificationAIPSREUSEDcanonical external
LEARNED_MODELClassification valueAIPSREUSEDcanonical external
SYMBOLIC_REASONERClassification valueAIPSREUSEDcanonical external
KNOWLEDGE_REASONERClassification valueAIPSREUSEDcanonical external
PLANNERClassification valueAIPSREUSEDcanonical external
INTELLIGENT_SEARCHClassification valueAIPSREUSEDcanonical external
OPTIMIZERClassification valueAIPSREUSEDcanonical external
HYBRID_INTELLIGENCE_ENGINEClassification valueAIPSREUSEDcanonical external
classifiedAsPrimitiveTypeRelationship TypeGEAIRMOWNEDcanonical
contributesToIntelligenceCapabilityRelationship TypeGEAIRMOWNED in M04reused by M05
Intelligence Mechanismeditorial umbrella termGEAIRM editorial usagenot a canonical model elementresolved - no standalone entity

AIModel and ModelProduct are referenced because AIPS gives them explicit relationships to AIPrimitive, but M05 does not redefine their Product or Asset semantics.


93. M05 Hard Distinctions​

The following distinctions are normative for subsequent slices:

AI Primitive != AI Primitive Type
AI Primitive Type != Intelligence Capability
AI Primitive Type != Execution Mode
AI Primitive Type != Intelligence Effect
AI Primitive Type != AI Product Type
AI Primitive Type != Product Family

LEARNED_MODEL != AIModel
LEARNED_MODEL != ModelProduct
SYMBOLIC_REASONER != ordinary deterministic routing
KNOWLEDGE_REASONER != knowledge resource
PLANNER != Planning != Plan
INTELLIGENT_SEARCH != ordinary retrieval
OPTIMIZER != Optimization
HYBRID_INTELLIGENCE_ENGINE != Composite Derived AI Product
HYBRID_INTELLIGENCE_ENGINE != Compound Product
HYBRID_INTELLIGENCE_ENGINE != multi-Agent system

AIAsset != automatically AIPrimitive
Product != automatically AIPrimitive
AIPrimitive != automatically AIProduct

As in M04, != denotes semantic non-equivalence rather than formal OWL disjointness unless executable formalization later explicitly authorizes stronger axioms.


94. M05 Normative Rules​

For traceability, M05 establishes the following rules.

RuleRequirement
M05-R01GEAIRM MUST preserve AIPS authority for AIPrimitive, AIPrimitiveType and the current AIPS primitive-type values.
M05-R02GEAIRM MUST NOT mint a redundant IntelligenceMechanism concept merely as an alias or superclass for AIPrimitive.
M05-R03AI Primitive and AI Primitive Type MUST remain distinct semantic constructs.
M05-R04Primitive-type classification MUST NOT be modelled as subclass inheritance unless AIPS explicitly establishes such an axiom.
M05-R05GEAIRM MUST NOT silently extend the AIPS AI Primitive Type controlled vocabulary.
M05-R06AIPS evidence-cardinality constraints MUST NOT be generalized into universal AI Primitive cardinality constraints without authority.
M05-R07GEAIRM MUST NOT broaden AIPS primitiveType from AI Primitive Evidence to AI Primitive resources.
M05-R08classifiedAsPrimitiveType MUST NOT imply Product status, AI Product qualification, capability, Agent identity, risk or architecture realization.
M05-R09LEARNED_MODEL, AIModel and ModelProduct MUST remain distinct.
M05-R10LEARNED_MODEL classification MUST NOT by itself determine Intelligence Capability.
M05-R11Ordinary deterministic rules or routing MUST NOT be classified as SYMBOLIC_REASONER solely because they manipulate conditions or symbols.
M05-R12Determinism MUST NOT by itself disqualify a mechanism from AI Primitive status.
M05-R13Knowledge resources and Knowledge Reasoner primitives MUST remain distinct; primitive status MUST NOT transfer through use/dependency.
M05-R14PLANNER, Planning and Plan MUST remain distinct primitive-type, capability and effect semantics.
M05-R15Ordinary retrieval or lookup MUST NOT be treated as INTELLIGENT_SEARCH solely because search occurs.
M05-R16Optimization technique alone MUST NOT establish the OPTIMIZER primitive type; qualifying intelligent behaviour is required.
M05-R17OPTIMIZER and Optimization MUST remain distinct primitive-type and capability dimensions.
M05-R18HYBRID_INTELLIGENCE_ENGINE MUST NOT be inferred solely from multiple primitives, Products, Agents or dependencies.
M05-R19Hybrid Intelligence Engine MUST remain distinct from Composite derivation, Compound Product and AI Product Type.
M05-R20AI Primitive Type MUST NOT deterministically imply a specific Intelligence Capability without an explicit relationship or authoritative rule.
M05-R21AI Primitive Type MUST NOT by itself determine Execution Mode or Intelligence Effect.
M05-R22GEAIRM MUST preserve authoritative multi-typing of AIModel and ModelProduct relative to AIPrimitive.
M05-R23Asset or Product status MUST NOT be used to infer AI Primitive status without authoritative semantics.
M05-R24M05 MUST NOT pre-empt M11 by introducing generic primitive-composition or dependency-transitivity rules.
M05-R25A resource MUST NOT inherit a dependency's AI Primitive Type solely through invocation, containment or dependency.

95. M05 Exit Criteria​

GEAIRM-M05 is complete when:

  • AIPS AI Primitive Type is integrated as a reused Classification;
  • all seven current AIPS primitive-type values are explicitly represented;
  • AI Primitive and AI Primitive Type are separated;
  • the AIPS evidence-scoped primitiveType property is not broadened;
  • a GEAIRM generic primitive-to-type relationship is defined conservatively;
  • Learned Model, AIModel and ModelProduct are separated;
  • Symbolic Reasoner is separated from ordinary deterministic automation;
  • Knowledge Reasoner is separated from knowledge resources;
  • Planner, Planning and Plan remain distinct;
  • Intelligent Search is separated from ordinary retrieval;
  • Optimizer is separated from generic optimization technique and Optimization capability;
  • Hybrid Intelligence Engine is separated from Product composition and multi-Agent topology;
  • primitive type remains orthogonal to Intelligence Capability, Execution Mode and Intelligence Effect;
  • authoritative multi-typing is preserved without converse inference;
  • dependency/type inheritance is prohibited;
  • detailed composition and topology semantics remain deferred to M11;
  • M01-M04 semantic authority, abstraction and inference rules remain intact.

M05 conclusion: these criteria are satisfied by this slice. GEAIRM now has a stable mechanism classification model on which M06 can define the Intelligence Capability, Execution Mode and Intelligence Effect relationships without confusing what intelligence is provided, what mechanism contributes it, how it is invoked, or what it produces or causes.


96. Forward Work​

The next slice is:

GEAIRM-M06 - Intelligence Capability, Execution & Effect Model

M06 will reconcile the AIPS Intelligence Capability, Execution Mode and Intelligence Effect dimensions at generic reference-model level. It will preserve their orthogonality, define any generic relationships needed outside AI Product Version scope, and maintain the M04/M05 separation among capability, primitive mechanism, invocation mode and resulting effect.


97. GEAIRM-M06 - Intelligence Capability, Execution & Effect Model​

97.1 Purpose of this slice​

M04 established the core distinction between Artificial Intelligence, Intelligence Capability and AI Primitive. M05 then elaborated the AI Primitive and mechanism classification model. M06 now completes the third part of the core behavioural frame by reconciling what intelligence is provided, how it is operated, and what it produces or causes.

M06 therefore integrates three AIPS-owned dimensions:

  • Intelligence Capability (IC) — what intelligence is provided;
  • Interaction / Execution Mode (MO) — how the intelligence is invoked or operated;
  • Intelligence Effect (IE) — what the intelligence produces or causes.

The AIPS ontology currently attaches these dimensions to AIProductVersion through Product-scoped properties. GEAIRM also needs to represent execution semantics outside a Product boundary without broadening those AIPS properties. M06 therefore introduces a minimal generic execution occurrence and conservative GEAIRM-owned relationships around it.

M06 answers five questions:

  1. How are Intelligence Capability, Execution Mode and Intelligence Effect represented at generic reference-model level?
  2. How can GEAIRM describe an AI execution occurrence without turning deployment architecture into reference-model semantics?
  3. How does an AI execution relate to AI Primitives and Intelligence Capabilities?
  4. How does execution mode differ from capability, primitive type, agency/autonomy and embodiment?
  5. How does an Intelligence Effect differ from an Intelligence Capability, output artifact, Product, decision authority or action authority?

98. Source Semantics and Authority Resolution​

M06 preserves the current AIPS ontology and taxonomy.

The existing AIPS baseline defines:

98.1 Intelligence Capability​

Canonical values:

  • Perception;
  • Classification;
  • Detection;
  • Prediction;
  • Generation;
  • RecommendationRanking;
  • Reasoning;
  • Planning;
  • Optimization;
  • Control.

M04 already integrated this dimension as AIPS-owned and REUSED.

98.2 Execution Mode​

AIPS ExecutionMode answers:

How is the intelligence invoked or operated?

Canonical AIPS values are:

  • Batch;
  • RealTimeEventDriven;
  • SynchronousAPI;
  • Interactive;
  • Embedded;
  • Continuous;
  • AgenticGoalDriven.

The existing AIPS property hasExecutionMode has domain AIProductVersion and range ExecutionMode.

98.3 Intelligence Effect​

AIPS IntelligenceEffect answers:

What does the intelligence produce or cause?

Canonical AIPS values are:

  • Score;
  • ClassificationEffect;
  • PredictionEffect;
  • DetectionAlert;
  • RecommendationEffect;
  • GeneratedContent;
  • ExplanationEvidence;
  • Decision;
  • Plan;
  • DigitalAction;
  • PhysicalAction.

The existing AIPS property hasIntelligenceEffect has domain AIProductVersion and range IntelligenceEffect.

The AIPS prose deliberately uses Effect rather than merely Output because intelligence may cause digital or physical action rather than only emit a passive artifact.

M06 does not alter those AIPS classes, values or Product-scoped properties.

M06-R01 - AIPS dimension primacy: GEAIRM MUST preserve AIPS authority for Intelligence Capability, Execution Mode, Intelligence Effect and their current controlled values.


99. Core Capability-Execution-Effect Distinction​

The M06 behavioural model is organized around three separate questions:

DimensionQuestionSemantic characterExample
Intelligence CapabilityWhat intelligence is provided or exercised?capability classificationDetection, Reasoning, Planning
Execution ModeHow is the intelligence invoked or operated?execution/interaction classificationBatch, Interactive, Continuous
Intelligence EffectWhat does the intelligence produce or cause?effect classificationScore, Plan, DigitalAction

These dimensions are orthogonal.

For example, a Reasoning capability may be exercised in Batch, Interactive or AgenticGoalDriven modes and may produce ExplanationEvidence, a RecommendationEffect, a Decision or another effect depending on the declared context.

Likewise, an Interactive mode does not reveal whether the intelligence capability is Generation, Detection, Planning or Control.

M06-R02 - Three-axis separation: Intelligence Capability, Execution Mode and Intelligence Effect MUST remain distinct semantic dimensions and MUST NOT be collapsed into one taxonomy.


100. AI Execution​

100.1 Semantic-gap resolution​

AIPS defines Execution Mode but does not define a generic reference-model entity representing an occurrence of AI being operated outside an AI Product Version boundary.

GEAIRM requires such an entity in order to:

  • preserve the AIPS Product-scoped hasExecutionMode and hasIntelligenceEffect properties;
  • describe execution of non-Product AI Primitives;
  • separate a capability's potential from an occurrence in which that capability is exercised;
  • relate execution mode and resulting effect without introducing platform or deployment architecture.

M06 therefore establishes a GEAIRM-owned Concept named AI Execution.

100.2 Canonical definition​

AI Execution is a context-bounded occurrence in which Artificial Intelligence is invoked, operated or enacted such that one or more Intelligence Capabilities are exercised through qualifying AI mechanisms and may produce or cause one or more Intelligence Effects.

AI Execution is an L1 semantic Concept. It is not:

  • a deployment;
  • a runtime platform;
  • an inference server;
  • an architecture building block;
  • an AI Product;
  • an Agent;
  • an Execution Mode value;
  • an Intelligence Effect.

100.3 Element record​

FieldValue
Preferred labelAI Execution
Element kindConcept
DefinitionA context-bounded occurrence in which Artificial Intelligence is invoked, operated or enacted such that one or more Intelligence Capabilities are exercised through qualifying AI mechanisms and may produce or cause one or more Intelligence Effects.
Semantic authorityGEAIRM
Authority identifierGEAIRM identifier to be minted during executable formalization after M17.
Representation dispositionOWNED
Normative statuscanonical
Source sliceM06
Constraints / invariantsSemantic occurrence only; not architecture/runtime topology, Product, Agent, mode or effect.

100.4 Why AI Execution is not architecture leakage​

GEAIRM needs to distinguish an AI capability from an occurrence in which that capability is exercised. This is semantic occurrence modelling, not a prescription of how or where execution must be hosted.

GEAIRA may later assign architectural responsibility for executing, serving, orchestrating or governing AI. M06 does not define such responsibilities.

M06-R03 - Execution occurrence boundary: AI Execution MUST be interpreted as a semantic occurrence and MUST NOT be treated as a runtime component, platform service, ABB or deployment topology.


101. AI Execution and AI Primitive​

M06 introduces a GEAIRM relationship for explicit participation of a qualifying AI Primitive in an AI Execution.

101.1 involvesAIPrimitive​

FieldValue
Preferred labelinvolves AI Primitive
DefinitionRelates an AI Execution to an AI Primitive that materially participates in that execution in the declared semantic context.
Semantic authorityGEAIRM
Subject kindsAI Execution
Object kindsaips:AIPrimitive
Directionalitydirected: AI Execution -> AI Primitive
Transitivitynot asserted
Symmetryfalse / not asserted
Inversenone normatively established by M06
Source sliceM06

involvesAIPrimitive does not assert:

  • Product dependency;
  • exclusive realization;
  • deployment co-location;
  • primitive ownership;
  • Agent identity;
  • primitive type;
  • direct or transitive Product qualification;
  • architecture topology.

M11 will define generic dependency and composition semantics. M06 therefore does not infer primitive involvement through arbitrary dependency paths.

M06-R04 - Primitive participation restraint: involvesAIPrimitive MUST NOT be interpreted as a generic dependency-transitivity rule or Product qualification path.


102. AI Execution and Intelligence Capability​

M04 established that an AI Primitive may contributeToIntelligenceCapability. M06 now distinguishes this capability contribution from an execution occurrence in which a capability is actually exercised.

102.1 exercisesIntelligenceCapability​

FieldValue
Preferred labelexercises Intelligence Capability
DefinitionRelates an AI Execution to an AIPS Intelligence Capability exercised during that execution in the declared semantic context.
Semantic authorityGEAIRM
Subject kindsAI Execution
Object kindsaips:IntelligenceCapability values
Directionalitydirected: AI Execution -> Intelligence Capability
Transitivitynot asserted
Symmetryfalse / not asserted
Inversenone normatively established by M06
Source sliceM06

The relationship distinguishes:

AI Primitive
contributesToIntelligenceCapability
-> capability potential/contribution

AI Execution
exercisesIntelligenceCapability
-> capability exercised in this occurrence/context

One does not imply the other without an explicit semantic assertion or authoritative rule.

For example, a primitive may be capable of contributing Reasoning and Planning, while one particular execution exercises only Reasoning.

M06-R05 - Potential/occurrence separation: A primitive's contribution to an Intelligence Capability MUST NOT be used to infer that every execution involving the primitive exercises that capability.


103. AI Execution and Execution Mode​

103.1 Execution Mode as reused Classification​

M06 integrates AIPS ExecutionMode as a reused Classification dimension.

FieldValue
Preferred labelInteraction / Execution Mode
Element kindClassification
DefinitionAIPS-controlled dimension answering how intelligence is invoked or operated.
Semantic authorityAIPS
Authority identifieraips:ExecutionMode
Representation dispositionREUSED
Normative statusreferenced/canonical through AIPS
Source sliceAIPS; integrated by GEAIRM M06
Canonical valuesBatch, RealTimeEventDriven, SynchronousAPI, Interactive, Embedded, Continuous, AgenticGoalDriven.

103.2 operatesInExecutionMode​

M06 introduces a generic relationship without widening AIPS hasExecutionMode.

FieldValue
Preferred labeloperates in Execution Mode
DefinitionRelates an AI Execution to an AIPS Execution Mode characterizing how that execution is invoked or operated in the declared semantic context.
Semantic authorityGEAIRM
Subject kindsAI Execution
Object kindsaips:ExecutionMode values
Directionalitydirected: AI Execution -> Execution Mode
Transitivitynot asserted
Symmetryfalse / not asserted
Inversenone normatively established by M06
Source sliceM06

Execution Mode is contextual rather than intrinsic to an AI Primitive. The same primitive may participate in different executions with different modes.

Likewise, an AI Product Version may advertise or govern one or more execution modes through AIPS while its concrete executions remain separate occurrences.

M06-R06 - Product-property preservation: GEAIRM MUST NOT broaden aips:hasExecutionMode beyond its AIPS AIProductVersion domain merely to describe generic AI execution.

M06-R07 - Mode contextuality: Execution Mode MUST NOT be treated as an intrinsic type of AI Primitive, AI Product, Agent or Intelligence Capability.


104. Execution Mode Semantic Boundaries​

104.1 Batch​

Batch describes invocation or operation over grouped/work-unit execution rather than a specific capability, primitive mechanism or effect.

Batch does not imply:

  • offline learning;
  • non-interactive capability in every other context;
  • lower risk;
  • absence of Agents;
  • absence of real-time dependencies.

104.2 RealTimeEventDriven​

RealTimeEventDriven describes event-driven or near-immediate invocation semantics. It does not imply Continuous operation, autonomous action, or any particular Intelligence Capability.

104.3 SynchronousAPI​

SynchronousAPI describes a request-response invocation mode. It does not establish an architecture product, API gateway, protocol, vendor or physical deployment pattern.

104.4 Interactive​

Interactive describes an execution mode involving iterative or interactive invocation. It does not by itself establish:

  • a human-in-the-loop control model;
  • Agent identity;
  • Assist or Recommend autonomy classification;
  • conversational AI;
  • a particular interface technology.

104.5 Embedded​

Embedded as an Execution Mode indicates intelligence operated as part of a containing system or component context. It is not equivalent to AIPS Embodiment.

Embedded Execution Mode != Embodied / Actuating

One concerns how intelligence is operated; the other concerns relationship to the physical world.

104.6 Continuous​

Continuous describes ongoing or continuously operating intelligence execution. It does not imply:

  • continuous learning;
  • retraining;
  • autonomous action;
  • physical embodiment;
  • closed-loop autonomy.

104.7 AgenticGoalDriven​

AgenticGoalDriven is an AIPS Execution Mode value. It indicates goal-driven/agentic operation at the AIPS mode dimension.

It does not by itself establish:

  • KivaNura Agent Ontology identity;
  • a specific Agent subtype;
  • Agent autonomy level;
  • AIPS Agency/Autonomy value;
  • AI Product Type;
  • autonomous decision/action authority.

Detailed reconciliation with Agent, Agent Autonomy and tool use is deferred to M08.

M06-R08 - Agentic-mode restraint: AgenticGoalDriven Execution Mode MUST NOT by itself establish Agent identity, Agent subtype, Agent Autonomy Level or AIPS Agency/Autonomy classification.

M06-R09 - Embedded/embodiment distinction: Embedded Execution Mode MUST remain distinct from AIPS Embodiment semantics.

M06-R10 - Continuous/learning distinction: Continuous Execution Mode MUST NOT be interpreted as continuous learning, retraining or self-modification unless separately asserted.


105. Intelligence Effect​

105.1 Intelligence Effect as reused Classification​

M06 integrates AIPS IntelligenceEffect as a reused Classification dimension.

FieldValue
Preferred labelIntelligence Effect
Element kindClassification
DefinitionAIPS-controlled dimension answering what the intelligence produces or causes.
Semantic authorityAIPS
Authority identifieraips:IntelligenceEffect
Representation dispositionREUSED
Normative statusreferenced/canonical through AIPS
Source sliceAIPS; integrated by GEAIRM M06
Canonical valuesScore, ClassificationEffect, PredictionEffect, DetectionAlert, RecommendationEffect, GeneratedContent, ExplanationEvidence, Decision, Plan, DigitalAction, PhysicalAction.

105.2 Effect is broader than passive output​

GEAIRM preserves the AIPS use of the term Effect rather than reducing the dimension to an output-artifact taxonomy.

An Intelligence Effect may represent:

  • a value, such as a Score;
  • a semantic result, such as a ClassificationEffect or PredictionEffect;
  • an alert;
  • a recommendation;
  • generated content;
  • explanation/evidence;
  • a Decision;
  • a Plan;
  • a DigitalAction;
  • a PhysicalAction.

The reference model therefore distinguishes:

Effect
may be informational
may be decisional
may be actional

without asserting that every effect is a persistent artifact, Data Asset, Information Product or Product.

M06-R11 - Effect/output distinction: Intelligence Effect MUST NOT be reduced to a passive output-artifact concept; digital and physical actions remain valid AIPS effects.


106. AI Execution and Intelligence Effect​

106.1 producesOrCausesIntelligenceEffect​

M06 introduces a generic relationship without changing AIPS hasIntelligenceEffect.

FieldValue
Preferred labelproduces or causes Intelligence Effect
DefinitionRelates an AI Execution to an AIPS Intelligence Effect that the execution produces or causes in the declared semantic context.
Semantic authorityGEAIRM
Subject kindsAI Execution
Object kindsaips:IntelligenceEffect values
Directionalitydirected: AI Execution -> Intelligence Effect
Transitivitynot asserted
Symmetryfalse / not asserted
Inversenone normatively established by M06
Source sliceM06

The relationship does not imply:

  • that the effect is the only effect of the execution;
  • that the effect persists as an artifact;
  • that the effect is a Product;
  • that the effect is an Information Product;
  • that a Decision is legally/finally authoritative;
  • that an Action is autonomous;
  • that the responsible Agent or Product owns the effect;
  • that the effect is attributable to one primitive exclusively.

M06-R12 - Product-property preservation for effects: GEAIRM MUST NOT broaden aips:hasIntelligenceEffect beyond its AIPS AIProductVersion domain merely to describe generic AI execution.

M06-R13 - Effect non-productization: An Intelligence Effect MUST NOT be treated as a Product, Information Product, Data Product or durable asset merely because it is produced or caused by AI.


107. Capability Does Not Determine Effect​

Capability and effect often have intuitive linguistic pairings, but M06 does not make them deterministic.

Examples of tempting but invalid universal mappings include:

Classification       -> ClassificationEffect
Prediction -> PredictionEffect
Detection -> DetectionAlert
RecommendationRanking-> RecommendationEffect
Generation -> GeneratedContent
Planning -> Plan
Control -> DigitalAction / PhysicalAction

Such pairings may be valid in a declared scenario, but they are not universal semantic equivalences.

A Detection capability might produce a Score and DetectionAlert. A Reasoning capability might produce ExplanationEvidence, a RecommendationEffect or a Plan. A Generation capability might produce content that is later interpreted as evidence, recommendation or another business artifact.

Therefore GEAIRM uses the AI Execution occurrence to relate exercised capability to produced effect rather than asserting a one-to-one capability-to-effect mapping.

M06-R14 - Capability/effect non-determinism: No Intelligence Capability value MUST be treated as universally determining a particular Intelligence Effect value merely because labels or common implementations correspond.


108. Execution Mode Does Not Determine Effect​

Execution Mode characterizes how intelligence is invoked or operated. It does not determine what effect occurs.

For example:

Interactive
may produce Score
may produce RecommendationEffect
may produce GeneratedContent
may produce Plan

Batch
may produce ClassificationEffect
may produce PredictionEffect
may produce GeneratedContent

AgenticGoalDriven
may produce Plan
may produce DigitalAction
may produce RecommendationEffect

These examples are illustrative and non-exhaustive.

Likewise, a PhysicalAction does not require one particular Execution Mode, and DigitalAction does not imply AgenticGoalDriven execution.

M06-R15 - Mode/effect orthogonality: Execution Mode MUST NOT be used to infer Intelligence Effect, and Intelligence Effect MUST NOT be used to infer Execution Mode, absent an explicit contextual assertion or authoritative rule.


109. Intelligence Effect vs Agency, Authority and Embodiment​

M06 integrates Intelligence Effect without pre-empting M08-M10.

109.1 Recommendation effect vs Recommend agency​

AIPS RecommendationEffect and AIPS Agency/Autonomy value Recommend answer different questions:

RecommendationEffect
what was produced

Recommend
what causal/decision authority is delegated

The same semantic label root does not establish equivalence.

109.2 Decision effect vs decision authority​

AIPS Decision is an Intelligence Effect value. M06 does not interpret this as final legal, operational or organizational decision authority.

M10 will reconcile Decision, recommendation, authorization and action semantics.

109.3 Digital/physical action vs autonomy​

DigitalAction or PhysicalAction indicates an effect that changes or acts upon a digital or physical environment. It does not by itself say whether the action was:

  • human-approved;
  • human-supervised;
  • delegated;
  • autonomous;
  • reversible;
  • safety-critical.

Agency/Autonomy remains a separate AIPS dimension.

109.4 PhysicalAction vs Embodiment​

A PhysicalAction effect is semantically related to the physical world, but M06 does not infer an AIPS Embodiment value automatically from one effect assertion. M09 will reconcile embodiment, environment and control-loop semantics.

M06-R16 - Effect/agency distinction: Intelligence Effect MUST remain distinct from AIPS Agency/Autonomy and MUST NOT by itself determine delegated authority.

M06-R17 - Decision-authority restraint: aips:Decision Intelligence Effect MUST NOT by itself establish final decision authority, authorization or organizational accountability.

M06-R18 - Action/autonomy restraint: DigitalAction and PhysicalAction MUST NOT by themselves establish autonomous action or a particular Agency/Autonomy value.

M06-R19 - Effect/embodiment restraint: PhysicalAction Intelligence Effect MUST NOT automatically determine an AIPS Embodiment classification before M09 establishes the applicable semantic relationship.


110. AI Execution Cardinality and Combination Semantics​

M06 imposes no universal single-valued constraint on Intelligence Capability, Execution Mode or Intelligence Effect at generic reference-model level.

An AI Execution may, where explicitly asserted:

  • exercise multiple Intelligence Capabilities;
  • operate under multiple applicable Execution Mode classifications;
  • produce or cause multiple Intelligence Effects;
  • involve multiple AI Primitives.

For example:

AI Execution E1
involvesAIPrimitive FraudDetector-X
exercisesIntelligenceCapability Detection
exercisesIntelligenceCapability Prediction
operatesInExecutionMode RealTimeEventDriven
producesOrCausesIntelligenceEffect Score
producesOrCausesIntelligenceEffect DetectionAlert
producesOrCausesIntelligenceEffect ExplanationEvidence

This does not imply that all executions involving FraudDetector-X have the same mode or effects.

GEAIRM also does not claim that every combination is semantically valid in every context. Context, authority, risk, embodiment and governance constraints are developed in later slices and external authorities.

M06-R20 - Multi-valued semantics: GEAIRM MUST NOT impose universal single-valued cardinality on Intelligence Capability, Execution Mode, Intelligence Effect or AI Primitive participation unless an applicable authoritative context explicitly requires it.


111. Invocation, Operation and Interaction Are Not Deployment Topology​

Execution Mode contains labels such as SynchronousAPI, Embedded and RealTimeEventDriven that may resemble implementation patterns. In AIPS they are semantic execution classifications, not infrastructure prescriptions.

Therefore GEAIRM MUST NOT infer from Execution Mode alone:

  • a specific API protocol;
  • synchronous transport technology;
  • event broker technology;
  • cloud service;
  • co-location;
  • microservice boundary;
  • model server;
  • edge device;
  • platform plane;
  • runtime product.

GEAIRA and concrete deployment specifications may map these semantic modes to architecture and technology.

M06-R21 - Mode/architecture separation: Execution Mode classifications MUST NOT be treated as GEAIRA ABBs, deployment products or technology selections.


112. Capability, Execution and Effect vs Product Semantics​

AIPS attaches IC, MO and IE to AIProductVersion because those dimensions describe a governed AI Product version. GEAIRM broadens only the semantic frame, not the AIPS Product qualification model.

The generic model therefore permits:

AI Product Version
hasIntelligenceCapability [AIPS]
hasExecutionMode [AIPS]
hasIntelligenceEffect [AIPS]

AI Execution
exercisesIntelligenceCapability [GEAIRM]
operatesInExecutionMode [GEAIRM]
producesOrCausesIntelligenceEffect [GEAIRM]

These are complementary levels.

The GEAIRM relationships MUST NOT be used to fabricate missing AI Product Version metadata, bypass AIPS qualification or infer Product Promise.

M06-R22 - Product/reference-model separation: Generic GEAIRM execution assertions MUST NOT substitute for required AIPS AI Product Version declarations or qualification evidence.


113. Capability, Mode and Effect vs Primitive Type​

M05 established that AI Primitive Type answers what kind of qualifying mechanism the primitive is. M06 now completes the orthogonal matrix:

AI Primitive Type
What kind of mechanism is involved?

Intelligence Capability
What kind of intelligence is exercised?

Execution Mode
How is it invoked or operated?

Intelligence Effect
What does it produce or cause?

None of the four dimensions universally determines another.

Illustratively:

SYMBOLIC_REASONER
+ Reasoning
+ Interactive
+ ExplanationEvidence

and:

LEARNED_MODEL
+ Detection
+ RealTimeEventDriven
+ Score + DetectionAlert

are valid-looking semantic combinations, but they are examples rather than universal mappings.

M06-R23 - Four-axis orthogonality: AI Primitive Type, Intelligence Capability, Execution Mode and Intelligence Effect MUST remain independent classification dimensions unless a narrower contextual rule is explicitly asserted.


114. Canonical M06 Reference View​

The M06 reference view is:

                         +----------------------+
| AI Execution |
| GEAIRM-owned |
+----------+-----------+
|
+-------------------------+-------------------------+
| | |
| | |
v v v
involvesAIPrimitive exercisesIntelligenceCapability operatesInExecutionMode
| | |
v v v
+-----------------+ +----------------------+ +----------------------+
| AI Primitive | | Intelligence | | Execution Mode |
| AIPS-owned | | Capability | | AIPS-owned |
+--------+--------+ | AIPS-owned | +----------------------+
| +----------------------+
|
contributesToIntelligenceCapability
|
+---------------------------> Intelligence Capability

AI Execution
|
+-- producesOrCausesIntelligenceEffect --> Intelligence Effect
AIPS-owned

The view intentionally does not show Product, Agent, Agency/Autonomy, Embodiment, Data/Knowledge dependencies or architecture/runtime components. Those concerns remain orthogonal or are assigned to later slices.


115. Illustrative Semantic Examples​

115.1 Real-time fraud assessment​

Execution E-Fraud-01
involvesAIPrimitive FraudDetector-v4.1
exercisesIntelligenceCapability Detection
exercisesIntelligenceCapability Prediction
operatesInExecutionMode RealTimeEventDriven
producesOrCausesIntelligenceEffect Score
producesOrCausesIntelligenceEffect DetectionAlert
producesOrCausesIntelligenceEffect ExplanationEvidence

No conclusion follows merely from this example that the primitive itself is an AI Product, that the execution is autonomous, or that the score is an Information Product.

115.2 Interactive reasoning​

Execution E-Reason-01
involvesAIPrimitive PolicyReasoner-X
exercisesIntelligenceCapability Reasoning
operatesInExecutionMode Interactive
producesOrCausesIntelligenceEffect ExplanationEvidence
producesOrCausesIntelligenceEffect RecommendationEffect

Interactive does not imply human approval or Agent identity.

115.3 Planner execution​

Execution E-Plan-01
involvesAIPrimitive Planner-X
exercisesIntelligenceCapability Planning
operatesInExecutionMode AgenticGoalDriven
producesOrCausesIntelligenceEffect Plan

This still does not assert that Planner-X is an Agent or that the Plan is executed autonomously.

115.4 Physical action​

Execution E-Physical-01
exercisesIntelligenceCapability Control
producesOrCausesIntelligenceEffect PhysicalAction

M06 does not infer an Embodiment classification, autonomy level or safety status. Those require separate assertions and later semantic reconciliation.

115.5 Continuous generation​

Execution E-Gen-01
exercisesIntelligenceCapability Generation
operatesInExecutionMode Continuous
producesOrCausesIntelligenceEffect GeneratedContent

Continuous does not imply continuous learning or retraining.


116. M06 Canonical Element Inventory​

M06 establishes or integrates the following model elements:

ElementM03 kindAuthorityDispositionM06 status
AI ExecutionConceptGEAIRMOWNEDcanonical
Intelligence CapabilityClassificationAIPSREUSEDcanonical external; established in M04
Execution ModeClassificationAIPSREUSEDcanonical external
BatchClassification valueAIPSREUSEDcanonical external
RealTimeEventDrivenClassification valueAIPSREUSEDcanonical external
SynchronousAPIClassification valueAIPSREUSEDcanonical external
InteractiveClassification valueAIPSREUSEDcanonical external
EmbeddedClassification valueAIPSREUSEDcanonical external
ContinuousClassification valueAIPSREUSEDcanonical external
AgenticGoalDrivenClassification valueAIPSREUSEDcanonical external
Intelligence EffectClassificationAIPSREUSEDcanonical external
ScoreClassification valueAIPSREUSEDcanonical external
ClassificationEffectClassification valueAIPSREUSEDcanonical external
PredictionEffectClassification valueAIPSREUSEDcanonical external
DetectionAlertClassification valueAIPSREUSEDcanonical external
RecommendationEffectClassification valueAIPSREUSEDcanonical external
GeneratedContentClassification valueAIPSREUSEDcanonical external
ExplanationEvidenceClassification valueAIPSREUSEDcanonical external
DecisionClassification valueAIPSREUSEDcanonical external; deeper reconciliation deferred to M10
PlanClassification valueAIPSREUSEDcanonical external
DigitalActionClassification valueAIPSREUSEDcanonical external
PhysicalActionClassification valueAIPSREUSEDcanonical external
involvesAIPrimitiveRelationship TypeGEAIRMOWNEDcanonical
exercisesIntelligenceCapabilityRelationship TypeGEAIRMOWNEDcanonical
operatesInExecutionModeRelationship TypeGEAIRMOWNEDcanonical
producesOrCausesIntelligenceEffectRelationship TypeGEAIRMOWNEDcanonical

117. M06 Hard Distinctions​

The following distinctions are normative for subsequent slices:

AI Execution != AI Primitive
AI Execution != AI Product
AI Execution != Agent
AI Execution != deployment/runtime component
AI Execution != Execution Mode
AI Execution != Intelligence Effect

Intelligence Capability != Execution Mode
Intelligence Capability != Intelligence Effect
Execution Mode != Intelligence Effect

Embedded Execution Mode != Embodiment
AgenticGoalDriven Execution Mode != Agent identity
AgenticGoalDriven Execution Mode != Agency/Autonomy
Continuous Execution Mode != continuous learning

RecommendationEffect != Recommend Agency/Autonomy
Decision Effect != decision authority
DigitalAction != AutonomousAction
PhysicalAction != AutonomousAction
PhysicalAction != automatically Embodiment classification

Classification capability != ClassificationEffect
Prediction capability != PredictionEffect
Detection capability != DetectionAlert
RecommendationRanking capability != RecommendationEffect
Generation capability != GeneratedContent
Planning capability != Plan effect
Control capability != DigitalAction / PhysicalAction effect

As in M04-M05, != denotes semantic non-equivalence rather than formal OWL disjointness unless later executable formalization explicitly authorizes stronger axioms.


118. M06 Normative Rules​

For traceability, M06 establishes the following rules.

RuleRequirement
M06-R01GEAIRM MUST preserve AIPS authority for Intelligence Capability, Execution Mode, Intelligence Effect and their controlled values.
M06-R02Intelligence Capability, Execution Mode and Intelligence Effect MUST remain distinct semantic dimensions.
M06-R03AI Execution MUST remain a semantic occurrence and MUST NOT become a runtime component, ABB, platform service or deployment topology.
M06-R04involvesAIPrimitive MUST NOT be interpreted as dependency transitivity, Product qualification or architecture topology.
M06-R05Primitive capability contribution MUST NOT imply that every execution involving that primitive exercises the same capability.
M06-R06GEAIRM MUST NOT broaden AIPS hasExecutionMode beyond its AIProductVersion domain.
M06-R07Execution Mode MUST NOT be treated as an intrinsic type of AI Primitive, Product, Agent or Intelligence Capability.
M06-R08AgenticGoalDriven MUST NOT by itself establish Agent identity, Agent subtype, Agent Autonomy Level or AIPS Agency/Autonomy classification.
M06-R09Embedded Execution Mode MUST remain distinct from AIPS Embodiment semantics.
M06-R10Continuous Execution Mode MUST NOT be interpreted as continuous learning, retraining or self-modification.
M06-R11Intelligence Effect MUST NOT be reduced to a passive output-artifact concept.
M06-R12GEAIRM MUST NOT broaden AIPS hasIntelligenceEffect beyond its AIProductVersion domain.
M06-R13Intelligence Effect MUST NOT automatically become Product, Information Product, Data Product or durable asset.
M06-R14Intelligence Capability MUST NOT universally determine Intelligence Effect through label similarity or common implementation.
M06-R15Execution Mode MUST NOT infer Intelligence Effect and Intelligence Effect MUST NOT infer Execution Mode absent explicit contextual semantics.
M06-R16Intelligence Effect MUST remain distinct from Agency/Autonomy and MUST NOT determine delegated authority.
M06-R17Decision Intelligence Effect MUST NOT by itself establish final decision authority, authorization or organizational accountability.
M06-R18DigitalAction and PhysicalAction MUST NOT by themselves establish AutonomousAction or another Agency/Autonomy value.
M06-R19PhysicalAction Intelligence Effect MUST NOT automatically determine an Embodiment classification before M09 reconciliation.
M06-R20GEAIRM MUST NOT impose universal single-valued cardinality on capability, mode, effect or primitive participation without applicable authority.
M06-R21Execution Mode classifications MUST NOT be treated as architecture building blocks, deployment products or technology selections.
M06-R22Generic GEAIRM execution assertions MUST NOT substitute for required AIPS AI Product Version metadata or qualification evidence.
M06-R23AI Primitive Type, Intelligence Capability, Execution Mode and Intelligence Effect MUST remain orthogonal classification dimensions unless a narrower contextual rule is explicitly asserted.
M06-R24The same AI Primitive MAY participate in different AI Executions with different capability, mode and effect assertions.
M06-R25M06 MUST NOT pre-empt M08-M10 by equating agentic mode, effect or action with Agent identity, delegated authority, decision authority or action authorization.

119. M06 Exit Criteria​

GEAIRM-M06 is complete when:

  • AIPS Intelligence Capability remains reused without redefinition;
  • AIPS Execution Mode and all seven current controlled values are integrated;
  • AIPS Intelligence Effect and all eleven current controlled values are integrated;
  • Product-scoped AIPS hasExecutionMode and hasIntelligenceEffect properties are not broadened;
  • a generic AI Execution occurrence exists without introducing architecture/runtime topology;
  • AI Execution can explicitly relate to AI Primitive, exercised capability, execution mode and resulting effect;
  • capability potential is distinguished from capability exercised in a particular execution;
  • Execution Mode is contextual rather than intrinsic to primitive or Product identity;
  • AgenticGoalDriven does not imply Agent identity or autonomy;
  • Embedded does not imply Embodiment;
  • Continuous does not imply learning/retraining;
  • Intelligence Effect remains broader than passive output;
  • effect does not imply Productization;
  • capability-to-effect and mode-to-effect mappings remain non-deterministic;
  • RecommendationEffect is separated from Recommend autonomy;
  • Decision effect is separated from decision authority;
  • Digital/Physical Action effects are separated from autonomous authority;
  • M09 retains ownership of Embodiment reconciliation;
  • M10 retains ownership of recommendation/decision/authorization/action reconciliation;
  • M11 retains ownership of dependency and composition semantics;
  • M01-M05 authority, abstraction and conservative inference rules remain intact.

M06 conclusion: these criteria are satisfied by this slice. GEAIRM now has a stable Capability-Execution-Effect model that can describe how AI mechanisms participate in concrete semantic execution occurrences without collapsing capability, invocation, effect, Product semantics, Agent semantics or architecture realization.


120. Forward Work​

The next slice is:

GEAIRM-M07 - Data, Knowledge, Context & Grounding Model

M07 will resolve the M01 semantic-gap candidates around Knowledge, Rule, Context, Prompt/Instruction, Grounding, Retrieval and Memory while aligning Data and Information to GEDRM/GEARM rather than redefining them. It will also determine how these concepts participate in AI Execution and intelligence mechanisms without confusing source material, context construction, grounding, retrieval or memory with AI Primitive identity.


121. GEAIRM-M07 - Data, Knowledge, Context & Grounding Model​

121.1 Purpose of this slice​

M04-M06 established the core semantic nucleus around Artificial Intelligence, AI Primitive, Intelligence Capability, AI Execution, Execution Mode and Intelligence Effect. M07 now establishes the semantic resources and contextual structures through which an AI Execution receives, interprets and grounds relevant material.

M07 resolves the M01 semantic-gap candidates around:

  • Knowledge;
  • Rule;
  • Context;
  • Prompt / Instruction;
  • Grounding;
  • Retrieval;
  • Memory.

It also integrates Data and Information as cross-domain references without redefining GEDRM or GEARM semantics.

M07 answers seven questions:

  1. How does GEAIRM reference Data and Information without becoming their semantic authority?
  2. What generic AI semantic construct is needed for knowledge used by AI?
  3. How are Rule, Prompt and Instruction distinguished?
  4. What is AI Context, and how does it differ from M03 Modelling Context and BPS Business Context?
  5. How is Grounding represented without turning it into an intrinsic resource type?
  6. How does Retrieval differ from Intelligent Search and Grounding?
  7. What is Memory, and how does it contribute to AI Context without becoming synonymous with Context or learning?

M07 does not define architecture for vector stores, prompt registries, memory services, knowledge platforms or retrieval systems. Those realization responsibilities belong to GEAIRA or implementation specifications.


122. Source Semantics and Authority Resolution​

The current AIPS baseline already establishes several important source semantics that M07 preserves:

  • AI Product constitution may include AI Primitives together with data, knowledge, rules, tools and controls;
  • AIPROD lineage treats Data Products / Data Sources as governed information used for training, inference, reasoning, grounding or evaluation;
  • AIPROD lineage groups knowledge bases, ontologies, rulesets, policies, prompts and other semantic artifacts as material lineage/dependency concerns;
  • AI Product input contracts may require Data Products, knowledge, sensor or upstream AI Product inputs;
  • AIPS distinguishes a Knowledge Reasoner mechanism from the knowledge resources it consumes;
  • AIPS distinguishes Intelligent Search from ordinary retrieval;
  • prompt libraries, knowledge bases and vector stores may be valuable assets or Products without thereby becoming AI Products;
  • the Agent Ontology mentions memory as a characteristic of some Agent types but does not establish a generic enterprise-AI Memory class.

M07 therefore resolves the M01 candidates as follows.

M01 candidateM07 resolutionAuthority / disposition
DataReference authoritative Data semantics; do not define a GEAIRM Data class.GEDRM / CROSS-DOMAIN-REFERENCE
Data ProductReference applicable Data/Product authorities.GEDRM + BPS/DPP / CROSS-DOMAIN-REFERENCE
InformationReference authoritative Information/Analytics semantics.GEARM / CROSS-DOMAIN-REFERENCE
Information ProductReference GEARM and applicable Product authority.GEARM + Product authority / CROSS-DOMAIN-REFERENCE
KnowledgeDo not create a universal metaphysical Knowledge entity; establish the narrower operational Concept Knowledge Resource.GEAIRM / OWNED for Knowledge Resource
RuleEstablish a generic Rule Concept.GEAIRM / OWNED
ContextEstablish the L1 Concept AI Context.GEAIRM / OWNED
PromptEstablish Prompt as a distinct AI interaction/configuration artifact Concept.GEAIRM / OWNED
InstructionEstablish Instruction as a directive Concept distinct from Prompt.GEAIRM / OWNED
GroundingDo not create a generic substance/entity called Grounding; represent grounding through a Grounding Source Role and groundsAIContext relationship.GEAIRM / OWNED Role + Relationship
RetrievalEstablish Retrieval as a semantic activity/occurrence Concept.GEAIRM / OWNED
MemoryEstablish Memory as a retained-state/resource Concept usable across executions/interactions.GEAIRM / OWNED

M07-R01 - Cross-domain authority preservation: GEAIRM MUST reference rather than redefine Data, Data Product, Information and Information Product semantics where those concepts are governed by GEDRM, GEARM, BPS/DPP or another applicable authority.

M07-R02 - Semantic minimality: M07 MUST prefer operationally precise enterprise-AI concepts such as Knowledge Resource and AI Context over unnecessarily broad philosophical entities such as universal Knowledge or Context superclasses.


123. Data and Information as Cross-Domain Resources​

Data and Information participate extensively in AI scenarios, but participation does not transfer semantic authority to GEAIRM.

GEAIRM may state, for example, that an AI Execution uses Data or Information as contextual material, or that a Data Product provides material used for grounding. Such statements describe the AI relationship to the external concept; they do not redefine what Data or Information intrinsically is.

M07 therefore introduces no geairm:Data, geairm:DataProduct, geairm:Information or geairm:InformationProduct shadow concepts.

The following distinctions remain controlling:

Data != AI Context
Information != AI Context
Data Product != AI Product
Information Product != AI Product
use by AI != AI Primitive status
use by AI != AI Product status

A Data or Information resource may participate in AI Context without becoming an intrinsic AI-domain type.

M07-R03 - No Data/Information shadow ontology: GEAIRM MUST NOT mint local substitutes for externally authoritative Data or Information concepts merely to place them in an AI context graph.

M07-R04 - Participation non-inheritance: Use of Data, Information, Data Products or Information Products by an AI Execution MUST NOT transfer AI Primitive, AI Product, Agent or intelligence-mechanism identity to those resources.


124. Knowledge Resource​

124.1 Why M07 does not create a universal Knowledge entity​

The current AIPS baseline uses knowledge as an important dependency category but generally refers to concrete things such as knowledge bases, ontologies, knowledge graphs and governed knowledge sources.

GEAIRM does not need a universal theory of knowledge in order to model enterprise AI. M07 therefore establishes the narrower operational Concept Knowledge Resource.

124.2 Canonical definition​

A Knowledge Resource is an identifiable resource that represents, organizes or exposes domain knowledge for use by an AI or other computational process, without itself implying an intelligence mechanism, Agent, Product or AI Product.

Representative examples include:

  • knowledge bases;
  • ontologies;
  • knowledge graphs;
  • curated semantic repositories;
  • governed domain knowledge collections.

A Knowledge Resource may be implemented as Data, a service, a graph, a document collection or another technical form. Those implementation forms do not alter the reference-model meaning.

124.3 Element record​

FieldValue
Preferred labelKnowledge Resource
Element kindConcept
DefinitionIdentifiable resource representing, organizing or exposing domain knowledge for computational use.
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM07
Constraints / invariantsNot synonymous with Knowledge Reasoner, AI Primitive, Data Product, Information Product or Product.

124.4 Knowledge Resource vs Knowledge Reasoner​

M05 established that a Knowledge Reasoner is an AI Primitive mechanism and is distinct from the knowledge resource over which it reasons.

Knowledge Resource
may be consumed by
Knowledge Reasoner

but

Knowledge Resource != Knowledge Reasoner

A Knowledge Resource does not acquire AI Primitive status because a Knowledge Reasoner uses it.

M07-R05 - Knowledge/mechanism separation: Knowledge Resource MUST remain distinct from AIPS KNOWLEDGE_REASONER and from any AI Primitive that consumes or reasons over it.

M07-R06 - Knowledge/Product restraint: Knowledge Resource status MUST NOT by itself establish Product, Knowledge Product, Data Product or AI Product status.


125. Rule​

125.1 Canonical definition​

A Rule is an explicit declarative, logical, conditional, procedural or constraint statement intended to influence derivation, selection, validation or behaviour within a defined semantic or operational context.

Rules may be consumed by symbolic reasoners, knowledge reasoners, planners, Agents, conventional software or governance mechanisms.

Rule is therefore a semantic resource, not a synonym for the mechanism that interprets or executes it.

125.2 Element record​

FieldValue
Preferred labelRule
Element kindConcept
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM07
Constraints / invariantsRule is not Symbolic Reasoner, AI Primitive, Policy, Control, Instruction or decision authority merely by being present.

125.3 Rule vs mechanism​

Rule
= semantic statement / resource

Symbolic Reasoner
= qualifying AI Primitive mechanism

A ruleset used by a Symbolic Reasoner does not become the reasoner. Conversely, ordinary rules may be executed by conventional deterministic software without creating AI semantics.

125.4 Rule vs Policy and Control​

M07 does not define the full Policy/Control model; that work remains primarily in M13 and applicable governance specifications.

For M07 purposes:

Rule != Policy
Rule != Control

A Rule may express or operationalize part of a Policy or Control, but equivalence is not implied.

M07-R07 - Rule/mechanism separation: A Rule MUST NOT be treated as a Symbolic Reasoner or other AI Primitive merely because an AI mechanism interprets it.

M07-R08 - Rule/governance distinction: Rule, Policy and Control MUST remain non-equivalent unless an applicable authority establishes a specific mapping.


126. AI Context​

126.1 Canonical definition​

AI Context is the bounded set of semantically relevant resources, state, conditions, instructions and other information made available to, or considered by, a particular AI Execution or interaction for a defined task or invocation.

AI Context is an L1 enterprise-AI Concept. It is deliberately distinct from:

  • M03 Modelling Context, which scopes the interpretation of model assertions;
  • BPS Business Context, which supports contextual Product-role interpretation;
  • architecture/runtime constructs such as context stores, vector databases or prompt-management services.

126.2 Element record​

FieldValue
Preferred labelAI Context
Element kindConcept
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM07
Constraints / invariantsOccurrence/task scoped; not identical to Memory, Prompt, Grounding, Data, Business Context or M03 Modelling Context.

126.3 Possible context resources​

An AI Context may include or reference, where applicable:

  • Data or Data Products;
  • Information or Information Products;
  • Knowledge Resources;
  • Rules;
  • Prompts;
  • Instructions;
  • Memory-derived state;
  • prior Intelligence Effects;
  • tool results;
  • environment observations;
  • user or system input;
  • upstream AI Product outputs/effects.

This list is illustrative and not a closed taxonomy.

126.4 Context is not identity​

An AI Primitive or AI Product may participate in many executions with different contexts. Context variation does not automatically change the intrinsic identity of the primitive or Product.

However, a materially different context may affect evaluation, risk, behaviour or fitness. Those implications are addressed in M13/M14 and applicable Product governance specifications.

M07-R09 - Context-level distinction: GEAIRM AI Context, M03 Modelling Context and BPS Business Context MUST remain distinct constructs.

M07-R10 - Context non-intrinsicality: AI Context MUST NOT be treated as an intrinsic type of AI Primitive, AI Product, Agent or Intelligence Capability.


127. AI Context Resource Role​

M07 defines AI Context Resource as a Role rather than a new intrinsic superclass.

AI Context Resource is the contextual Role played by a resource when that resource is included in, referenced by or otherwise made semantically available within an AI Context.

Any suitable resource may play this Role, subject to its own semantic authority.

Examples include a Data Product, Knowledge Resource, Rule, Prompt, Memory item, tool result or environment observation.

This modelling choice avoids a shadow upper ontology in which every context participant is forced under a GEAIRM-owned ContextResource superclass.

M07-R11 - Context-resource role discipline: AI Context Resource MUST be treated as a contextual Role, not as an intrinsic superclass that changes the identity or authority of the participating resource.


128. hasAIContext​

M07 introduces a canonical relationship from AI Execution to AI Context.

FieldValue
Preferred labelhas AI Context
DefinitionRelates an AI Execution to an AI Context that is semantically applicable to that execution or interaction.
Semantic authorityGEAIRM
Subject kindsAI Execution
Object kindsAI Context
DirectionalityAI Execution -> AI Context
Transitivitynot asserted
Functional cardinalitynot asserted
Inversenone normatively established by M07
Source sliceM07

An execution may be represented with one or more applicable context structures, and M07 does not require a universal exact cardinality.

hasAIContext does not mean that every resource related to the context was actually used by every primitive in the execution.

M07-R12 - Execution/context cardinality restraint: hasAIContext MUST NOT be interpreted as universally single-valued or as proof that every context resource materially influenced the execution.


129. includesContextResource​

M07 introduces a relationship from AI Context to a resource playing the AI Context Resource Role.

FieldValue
Preferred labelincludes context resource
DefinitionRelates an AI Context to a resource that is included, referenced or made semantically available within that context.
Semantic authorityGEAIRM
Subject kindsAI Context
Object kindsany suitable Concept/resource under its applicable authority
DirectionalityAI Context -> resource
Transitivitynot asserted
Inversenone normatively established by M07
Source sliceM07

The relationship does not transfer semantic authority or intrinsic type to the resource.

M07-R13 - Context inclusion restraint: includesContextResource MUST NOT imply grounding, material use, truth, authorization, Product status, AI Primitive status or semantic ownership transfer.


130. Instruction​

130.1 Canonical definition​

An Instruction is an explicit directive, constraint or task-oriented statement intended to influence the behaviour, processing or response of an AI Execution, Agent or computational mechanism.

Instruction is a semantic directive. It is not automatically authorization, policy or control.

An Instruction may be conveyed through a Prompt, configuration, API field, policy-derived artifact, workflow, Agent state or another mechanism.

130.2 Element record​

FieldValue
Preferred labelInstruction
Element kindConcept
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM07
Constraints / invariantsInstruction is not Prompt, Policy, Rule, authority delegation or governance Control by default.

M07-R14 - Instruction/authority distinction: An Instruction MUST NOT by itself establish permission, delegated authority, legal authority or authorization to perform an action.

M07-R15 - Instruction/governance distinction: Instruction MUST remain distinct from Policy, Control and Rule unless an explicit semantic mapping is established.


131. Prompt​

131.1 Canonical definition​

A Prompt is an interaction or configuration artifact that presents input content to an AI mechanism or AI Execution and may carry one or more Instructions, examples, contextual material, questions or other task-relevant content.

Prompt is deliberately broader than Instruction. A prompt may contain an Instruction, but may also contain user content, demonstrations, source material or context without being reducible to a directive.

131.2 Element record​

FieldValue
Preferred labelPrompt
Element kindConcept
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM07
Constraints / invariantsPrompt is not intrinsically an Instruction, AI Primitive, AI Product, AI Context or authorization object.

131.3 Prompt vs Product/Asset​

A prompt or prompt library may itself be managed as an asset or even productized under an applicable Product ontology. M07 does not determine that Product status.

A prompt's participation in an AI Context likewise does not make the prompt an AI Product or AI Primitive.

M07-R16 - Prompt/instruction distinction: Prompt and Instruction MUST remain distinct; a Prompt MAY contain or convey Instructions but MUST NOT be treated as synonymous with Instruction.

M07-R17 - Prompt/Product restraint: Prompt usage in AI MUST NOT by itself establish Product, AI Product or AI Primitive status.


132. containsInstruction​

M07 introduces a relationship that allows a Prompt to explicitly carry an Instruction.

FieldValue
Preferred labelcontains instruction
DefinitionRelates a Prompt to an Instruction explicitly represented within or conveyed by that Prompt.
Semantic authorityGEAIRM
Subject kindsPrompt
Object kindsInstruction
DirectionalityPrompt -> Instruction
Transitivitynot asserted
Functional cardinalitynot asserted
Inversenone normatively established by M07
Source sliceM07

Not every Prompt is required to contain a separately modelled Instruction, and an Instruction may be conveyed without a Prompt.

M07-R18 - Prompt/instruction cardinality restraint: containsInstruction MUST NOT imply that every Prompt contains exactly one Instruction or that every Instruction must be represented through a Prompt.


133. Grounding Source Role and Grounding Relationship​

133.1 Grounding is contextual participation, not intrinsic identity​

M07 does not create a universal Grounding substance or artifact class.

Instead, Grounding Source is a Role played by a resource when that resource is used as external, reference, evidential or authoritative material to constrain, substantiate or anchor interpretation or output within an AI Context.

A resource can play Grounding Source in one execution and not in another.

Possible Grounding Sources include:

  • Data or Data Products;
  • Information or Information Products;
  • Knowledge Resources;
  • documents;
  • governed rules or policy artifacts;
  • tool/service results;
  • environment observations;
  • other evidence-bearing resources.

Grounding Source does not imply that the source is infallible, authoritative in every domain, or sufficient to guarantee correctness.

133.2 groundsAIContext​

FieldValue
Preferred labelgrounds AI Context
DefinitionRelates a resource playing the Grounding Source Role to an AI Context whose interpretation or output is intended to be constrained, substantiated or anchored by that resource.
Semantic authorityGEAIRM
Subject kindsresource under its applicable authority
Object kindsAI Context
Directionalityresource -> AI Context
Transitivitynot asserted
Inversenone normatively established by M07
Source sliceM07

Grounding is therefore a role/relationship semantics, not an intrinsic type of Data, Knowledge Resource, document or Product.

M07-R19 - Grounding-role discipline: Grounding Source MUST remain a contextual Role and MUST NOT become an intrinsic resource type solely because the resource grounds an AI Context.

M07-R20 - Grounding/truth restraint: groundsAIContext MUST NOT be interpreted as proof that the grounding source is correct, complete, authoritative for all purposes, or sufficient to guarantee a correct Intelligence Effect.


134. Retrieval​

134.1 Canonical definition​

Retrieval is a semantic activity in which one or more resources are located, selected or obtained from an accessible source for potential use by a subsequent process, including context construction or grounding.

Retrieval is not inherently Artificial Intelligence.

It may be implemented through:

  • deterministic lookup;
  • database query;
  • keyword search;
  • vector similarity;
  • indexed document retrieval;
  • a qualifying Intelligent Search AI Primitive;
  • a hybrid mechanism.

The implementation does not change the generic Retrieval semantics.

134.2 Element record​

FieldValue
Preferred labelRetrieval
Element kindConcept representing an activity/occurrence
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM07
Constraints / invariantsRetrieval is not automatically Intelligent Search, Grounding, AI Execution, AI Primitive or Agent action.

M05 already established:

Retrieval != automatically INTELLIGENT_SEARCH

M07 makes the relationship explicit: a Retrieval activity may be performed using a primitive classified INTELLIGENT_SEARCH, but ordinary retrieval can occur without any qualifying AI Primitive.

134.4 Retrieval vs Grounding​

Retrieval and Grounding answer different questions:

Retrieval
How was a resource located or obtained?

Grounding
What semantic role does a source play in anchoring the AI Context?

A resource may be retrieved but never used for grounding. A grounding source may already be supplied without retrieval.

M07-R21 - Retrieval/AI distinction: Retrieval MUST NOT by itself establish Artificial Intelligence or AI Primitive status.

M07-R22 - Retrieval/search distinction: Retrieval MUST NOT be treated as INTELLIGENT_SEARCH unless the mechanism itself satisfies the applicable AIPS AI Primitive semantics and is explicitly classified accordingly.

M07-R23 - Retrieval/grounding distinction: Retrieval and Grounding MUST remain distinct; obtaining a resource MUST NOT by itself establish that the resource grounded the AI Context.


135. Retrieval Relationships​

135.1 retrievesResource​

FieldValue
Preferred labelretrieves resource
DefinitionRelates a Retrieval activity to a resource located, selected or obtained by that activity.
Semantic authorityGEAIRM
Subject kindsRetrieval
Object kindsresource under its applicable authority
DirectionalityRetrieval -> resource
Transitivitynot asserted
Source sliceM07

135.2 suppliesToAIContext​

FieldValue
Preferred labelsupplies to AI Context
DefinitionRelates a Retrieval activity to an AI Context to which retrieved material is made available.
Semantic authorityGEAIRM
Subject kindsRetrieval
Object kindsAI Context
DirectionalityRetrieval -> AI Context
Transitivitynot asserted
Source sliceM07

suppliesToAIContext does not imply that every retrieved resource becomes a Grounding Source or materially influences the AI Execution.

M07-R24 - Retrieval/context restraint: Supplying retrieved material to AI Context MUST NOT imply that the material was used, accepted, grounded, correct or causally influential.


136. Memory​

136.1 Canonical definition​

Memory is a retained or retrievable state/resource that preserves information from prior or ongoing interactions, executions or observations so that it may influence a later AI Context or behaviour.

Memory is intentionally mechanism-neutral. It may be realized using conversational state, event history, structured records, embeddings, summaries, working state or other implementation forms.

136.2 Element record​

FieldValue
Preferred labelMemory
Element kindConcept
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM07
Constraints / invariantsMemory is not AI Context itself, not necessarily Knowledge Resource, not learning/retraining, not Agent identity, and not model parameters by default.

136.3 Memory vs AI Context​

Memory is persistent or retainable semantic state; AI Context is the bounded material applicable to a particular execution or interaction.

Therefore:

Memory
may contribute to
AI Context

but

Memory != AI Context

The same Memory resource may contribute to many later contexts, and an AI Context may be constructed without persistent Memory.

136.4 Memory vs learning​

Retaining interaction state does not imply that an AI model learned, retrained, fine-tuned or modified its parameters.

Likewise, AIPS Continuous Execution Mode does not imply Memory, and Memory does not imply Continuous mode.

136.5 Memory vs Agent identity​

Some Agent Ontology classes mention memory as a characteristic. M07 does not infer Agent identity from Memory use, nor does it change the Agent Ontology taxonomy.

M07-R25 - Memory/context distinction: Memory MUST remain distinct from AI Context; Memory MAY contribute to an AI Context but is not synonymous with the context itself.

M07-R26 - Memory/learning distinction: Memory retention MUST NOT by itself imply model learning, retraining, adaptation or parameter change.

M07-R27 - Memory/Agent restraint: Use of Memory MUST NOT by itself establish Agent identity, Agent subtype or autonomy level.


137. memoryContributesToAIContext​

FieldValue
Preferred labelmemory contributes to AI Context
DefinitionRelates Memory to an AI Context when retained or retrievable state from that Memory is made available as part of the context.
Semantic authorityGEAIRM
Subject kindsMemory
Object kindsAI Context
DirectionalityMemory -> AI Context
Transitivitynot asserted
Functional cardinalitynot asserted
Source sliceM07

The relationship does not imply that all Memory content is loaded into the context or that the context faithfully reproduces the full Memory resource.

M07-R28 - Memory contribution restraint: memoryContributesToAIContext MUST NOT be interpreted as complete Memory inclusion, semantic equivalence or proof that all retained state influenced the execution.


138. AI Context Assembly and Grounding Semantics​

M07 deliberately separates several commonly collapsed concerns.

Source resource
|
| may be located by
v
Retrieval
|
| supplies material to
v
AI Context
^
|
| may also receive
+---- Prompt / Instruction
+---- Memory-derived state
+---- Data / Information
+---- Knowledge Resource / Rule
+---- Tool result / environment observation

A resource within the context
|
| may play Grounding Source Role
v
groundsAIContext

AI Execution
|
+---- hasAIContext ----> AI Context

This is not a mandatory pipeline. In particular:

  • context material may be supplied without Retrieval;
  • Retrieval may occur without Grounding;
  • Grounding may use a pre-existing resource;
  • a Prompt may contribute context without containing an Instruction;
  • Memory may be absent;
  • a Rule may constrain behaviour without being a Grounding Source;
  • not every context resource materially affects the result.

M07-R29 - No forced pipeline: GEAIRM MUST NOT require Retrieval -> Grounding -> Context -> Execution as a universal sequence; these are separable semantic relationships that may be present in different combinations.


139. RAG and Grounded Generation as an Illustrative Pattern​

Retrieval-Augmented Generation (RAG) is useful for stress-testing the M07 distinctions but is not itself introduced as a new intrinsic Product or AI Primitive type.

An illustrative pattern is:

Retrieval R1
retrievesResource DocumentSet-D
suppliesToAIContext Context-C

DocumentSet-D
groundsAIContext Context-C

Execution E1
hasAIContext Context-C
exercisesIntelligenceCapability Generation
involvesAIPrimitive Generator-M
producesOrCausesIntelligenceEffect GeneratedContent

Possible variations include a qualifying Intelligent Search primitive performing the Retrieval, a Knowledge Resource acting as the Grounding Source, or Memory contributing additional context.

The pattern does not imply:

  • Retrieval is itself AI;
  • the retrieved corpus is an AI Primitive;
  • every RAG pipeline is an AI Product;
  • grounding guarantees factual correctness;
  • the generator is necessarily an Agent;
  • the vector store or knowledge base becomes an AI Product.

M07-R30 - RAG non-collapse: RAG MUST be treated as a pattern involving separable retrieval, context, grounding and intelligence-mechanism semantics; it MUST NOT be used as an intrinsic shortcut for AI Primitive Type, AI Product Type or Product qualification.


140. Prompt, Rule, Instruction and Policy Boundaries​

These concepts are often collapsed in implementations, but GEAIRM keeps them separate:

Prompt
interaction/configuration artifact

Instruction
directive intended to influence behaviour

Rule
declarative/logical/conditional/constraint statement

Policy
governance or organizational statement owned by applicable governance semantics

Control
governed mechanism or obligation used to constrain/assure behaviour

A Prompt may contain an Instruction. An Instruction may reflect a Policy. A Rule may operationalize a Policy. A Control may enforce a Rule. None of these relationships establishes equivalence.

Detailed Policy/Control semantics remain deferred to M13 and external governance authorities.

M07-R31 - Semantic boundary preservation: Prompt, Instruction, Rule, Policy and Control MUST remain non-equivalent unless an explicit mapping is established for a defined context.


141. Context and Grounding Do Not Imply Authority​

M07 distinguishes semantic availability from authority.

A resource may be included in AI Context or play a Grounding Source Role without being the authoritative source for the represented subject matter.

Likewise, an Instruction may be present in context without being authorized, and a retrieved resource may be relevant without being trusted.

Authority, entitlement, policy, safety and action authorization are separate concerns developed in M08, M10, M13 and applicable external specifications.

M07-R32 - Context/authority restraint: Inclusion in AI Context, Prompt, Memory, Retrieval or Grounding MUST NOT by itself establish authority, entitlement, trust, permission or authorization.


142. Context Resources Do Not Become AI Primitives​

M07 establishes a general non-inheritance rule for context resources.

Data
Information
Knowledge Resource
Rule
Prompt
Instruction
Memory
Grounding Source
Retrieved Resource

do not become

AI Primitive

merely because an AI Execution depends on or uses them.

A resource may independently qualify as an AI Primitive only if the applicable AIPS AI Primitive semantics are satisfied in its own right.

For example, an ordinary vector store or prompt library does not become an AI Primitive merely because it is critical to a generative AI system.

M07-R33 - Context/primitive non-inheritance: Participation as context, grounding, retrieval input, Memory or Instruction MUST NOT transfer AI Primitive identity to the participating resource.


143. Canonical M07 Reference View​

                           +-------------------+
| AI Execution |
| GEAIRM-owned |
+---------+---------+
|
hasAIContext
|
v
+-------------------+
| AI Context |
| GEAIRM-owned |
+----+-----+-----+--+
| | |
includes | | | memoryContributesToAIContext
context resource| | |
v | v
+---------------+ | +--------+
| resource in | | | Memory |
| Context Role | | +--------+
+-------+-------+ |
^ |
| |
groundsAIContext |
| |
+-------+---------+--------+
| Grounding Source Role |
+-------------------------+

Prompt ---- containsInstruction ----> Instruction
\
\ may participate in AI Context

Knowledge Resource / Rule / Data / Information / Tool result / Environment observation
may participate as AI Context Resources and, where explicitly asserted,
may play Grounding Source Role.

Retrieval ---- retrievesResource ----> resource
Retrieval ---- suppliesToAIContext --> AI Context

The diagram is conceptual. It does not prescribe platform components, storage technologies, RAG architectures, memory services or registry design.


144. Illustrative Semantic Examples​

144.1 Knowledge-grounded reasoning​

Knowledge Resource KR-Policy
groundsAIContext Context-1

Execution E-Reason
hasAIContext Context-1
involvesAIPrimitive Reasoner-X
exercisesIntelligenceCapability Reasoning

The Knowledge Resource is not the Knowledge Reasoner.

144.2 Rule-driven symbolic reasoning​

RuleSet-R
included as context resource in Context-2

Execution E-Symbolic
hasAIContext Context-2
involvesAIPrimitive SymbolicReasoner-X

RuleSet-R does not inherit AI Primitive status from SymbolicReasoner-X.

144.3 Prompt with instruction​

Prompt P1
containsInstruction I1

Context-C
includesContextResource P1

Execution E1
hasAIContext Context-C

The Instruction does not by itself authorize an action.

144.4 Memory-assisted interaction​

Memory M1
memoryContributesToAIContext Context-C2

Execution E2
hasAIContext Context-C2
operatesInExecutionMode Interactive

No learning or Agent identity follows from Memory use.

144.5 Ordinary retrieval​

Retrieval R1
retrievesResource Document-D
suppliesToAIContext Context-C3

Nothing in this statement says the Retrieval is Intelligent Search or Artificial Intelligence.


145. M07 Canonical Element Inventory​

ElementM03 kindAuthorityDispositionM07 status
DataConcept under external authorityGEDRMREFERENCEDcanonical external reference
Data ProductProduct / Data concept under external authoritiesGEDRM + BPS/DPPREFERENCEDcanonical external reference
InformationConcept under external authorityGEARMREFERENCEDcanonical external reference
Information ProductProduct / Information concept under external authoritiesGEARM + Product authorityREFERENCEDcanonical external reference
Knowledgeeditorial umbrellano standalone M07 entitynot canonicalresolved via Knowledge Resource
Knowledge ResourceConceptGEAIRMOWNEDcanonical
RuleConceptGEAIRMOWNEDcanonical
AI ContextConceptGEAIRMOWNEDcanonical
AI Context ResourceRoleGEAIRMOWNEDcanonical
InstructionConceptGEAIRMOWNEDcanonical
PromptConceptGEAIRMOWNEDcanonical
Grounding SourceRoleGEAIRMOWNEDcanonical
Groundingrole/relationship semanticsGEAIRMno standalone Conceptresolved
RetrievalConcept/activityGEAIRMOWNEDcanonical
MemoryConceptGEAIRMOWNEDcanonical
hasAIContextRelationship TypeGEAIRMOWNEDcanonical
includesContextResourceRelationship TypeGEAIRMOWNEDcanonical
containsInstructionRelationship TypeGEAIRMOWNEDcanonical
groundsAIContextRelationship TypeGEAIRMOWNEDcanonical
retrievesResourceRelationship TypeGEAIRMOWNEDcanonical
suppliesToAIContextRelationship TypeGEAIRMOWNEDcanonical
memoryContributesToAIContextRelationship TypeGEAIRMOWNEDcanonical

No executable GEAIRM identifiers are minted in M07. Stable machine identifiers remain deferred to A1-A8 after M17.


146. M07 Hard Distinctions​

The following distinctions are normative for subsequent slices:

Data != AI Context
Information != AI Context
Knowledge Resource != Knowledge Reasoner
Knowledge Resource != automatically Product
Rule != Symbolic Reasoner
Rule != Policy
Rule != Control

AI Context != M03 Modelling Context
AI Context != BPS Business Context
AI Context != Memory
AI Context != Prompt
AI Context != Grounding

Prompt != Instruction
Prompt != AI Primitive
Instruction != authorization
Instruction != Policy
Instruction != Rule

Grounding Source Role != intrinsic resource type
Grounding != Retrieval
Grounding != correctness guarantee
Retrieval != Intelligent Search
Retrieval != Artificial Intelligence

Memory != AI Context
Memory != learning/retraining
Memory != Agent identity

context participation != AI Primitive status
context participation != AI Product status

As in previous slices, != denotes semantic non-equivalence rather than formal OWL disjointness unless later executable formalization explicitly authorizes stronger axioms.


147. M07 Normative Rules​

RuleRequirement
M07-R01GEAIRM MUST reference rather than redefine authoritative Data, Data Product, Information and Information Product semantics.
M07-R02M07 MUST prefer operationally precise concepts over unnecessary universal philosophical entities where the narrower model is sufficient.
M07-R03GEAIRM MUST NOT create shadow Data or Information concepts merely for AI integration.
M07-R04Participation of Data/Information resources in AI MUST NOT transfer AI Primitive, AI Product or Agent identity.
M07-R05Knowledge Resource MUST remain distinct from Knowledge Reasoner and other AI Primitives.
M07-R06Knowledge Resource status MUST NOT by itself establish Product or AI Product status.
M07-R07Rule MUST NOT be treated as a Symbolic Reasoner or AI Primitive merely because a mechanism interprets it.
M07-R08Rule, Policy and Control MUST remain non-equivalent absent explicit mapping.
M07-R09AI Context, M03 Modelling Context and BPS Business Context MUST remain distinct.
M07-R10AI Context MUST NOT become an intrinsic type of primitive, Product, Agent or capability.
M07-R11AI Context Resource MUST be a contextual Role rather than an intrinsic superclass.
M07-R12hasAIContext MUST NOT imply universal single-valued cardinality or material influence by every context resource.
M07-R13includesContextResource MUST NOT imply grounding, truth, authorization, Product status, AI Primitive status or authority transfer.
M07-R14Instruction MUST NOT by itself establish permission, delegated authority or authorization.
M07-R15Instruction MUST remain distinct from Policy, Control and Rule absent explicit mapping.
M07-R16Prompt and Instruction MUST remain distinct; a Prompt MAY contain or convey Instructions.
M07-R17Prompt participation in AI MUST NOT by itself establish Product, AI Product or AI Primitive status.
M07-R18containsInstruction MUST NOT imply universal one-to-one Prompt/Instruction cardinality.
M07-R19Grounding Source MUST remain a contextual Role rather than an intrinsic resource type.
M07-R20Grounding MUST NOT be interpreted as proof of correctness, completeness or universal source authority.
M07-R21Retrieval MUST NOT by itself establish Artificial Intelligence or AI Primitive status.
M07-R22Retrieval MUST NOT be treated as Intelligent Search unless the qualifying mechanism semantics are explicitly satisfied.
M07-R23Retrieval and Grounding MUST remain distinct semantic concerns.
M07-R24Material supplied by Retrieval to AI Context MUST NOT be assumed used, accepted, grounded, correct or causally influential.
M07-R25Memory MUST remain distinct from AI Context.
M07-R26Memory retention MUST NOT by itself imply learning, retraining, adaptation or parameter change.
M07-R27Memory use MUST NOT by itself establish Agent identity, Agent subtype or autonomy.
M07-R28memoryContributesToAIContext MUST NOT imply complete Memory inclusion or full causal influence.
M07-R29Retrieval, Grounding, Context and AI Execution MUST NOT be forced into a universal mandatory pipeline.
M07-R30RAG MUST remain a compositional pattern and MUST NOT become an intrinsic shortcut for primitive type, AI Product Type or Product qualification.
M07-R31Prompt, Instruction, Rule, Policy and Control MUST remain semantically distinct unless explicitly mapped.
M07-R32Context, Prompt, Memory, Retrieval or Grounding participation MUST NOT by itself establish authority, trust, entitlement or authorization.
M07-R33Context/grounding/retrieval participation MUST NOT transfer AI Primitive identity to the participating resource.

148. M07 Exit Criteria​

GEAIRM-M07 is complete when:

  • Data and Data Product remain external cross-domain references;
  • Information and Information Product remain external cross-domain references;
  • a universal metaphysical Knowledge entity is avoided while Knowledge Resource is explicitly modelled;
  • Knowledge Resource is separated from Knowledge Reasoner;
  • Rule is modelled without becoming Symbolic Reasoner, Policy or Control;
  • AI Context is defined and separated from M03 Modelling Context and BPS Business Context;
  • AI Context Resource is represented as a Role rather than an intrinsic superclass;
  • Prompt and Instruction are defined and kept distinct;
  • Instruction is separated from authority and authorization;
  • Grounding is represented through role/relationship semantics rather than an intrinsic resource class;
  • Grounding is separated from correctness and source-authority claims;
  • Retrieval is defined independently of Intelligent Search and AI Primitive status;
  • Retrieval is separated from Grounding;
  • Memory is defined and separated from AI Context, learning and Agent identity;
  • AI Execution can be related to AI Context without changing M06 execution semantics;
  • context resources can participate without inheriting AI Primitive or AI Product status;
  • the RAG pattern can be expressed compositionally without introducing a special Product or primitive classification;
  • no architecture/runtime technology is prescribed;
  • M01-M06 semantic authority, abstraction and conservative inference rules remain intact.

M07 conclusion: these criteria are satisfied by this slice. GEAIRM now has a stable semantic model for the resources and contextual structures through which enterprise AI receives Data, Information, Knowledge Resources, Rules, Prompts, Instructions, retrieved material and Memory, while keeping Grounding, Retrieval, Context and intelligence mechanisms semantically distinct.


149. Forward Work​

The next slice is:

GEAIRM-M08 - Agent, Agency, Autonomy & Tool Model

M08 will reconcile AIPS Agency/Autonomy with the KivaNura Agent Ontology, preserve Agent semantic authority, resolve the generic Tool semantic gap, and define the relationships among Agent, AI Primitive, Intelligence Capability, AI Execution, Tool use, delegation, oversight and autonomy without collapsing Agent identity into agentic execution mode or AI Product status.​

150. GEAIRM-M08 - Agent, Agency, Autonomy & Tool Model​

150.1 Purpose of this slice​

M07 established the contextual resources through which enterprise AI receives Data, Information, Knowledge Resources, Rules, Prompts, Instructions, retrieved material and Memory. M08 now establishes how Agents participate in enterprise AI and reconciles the distinct semantics of Agent identity, operational autonomy, Product-level Agency/Autonomy, control/oversight, delegated authority and Tool use.

M08 answers seven questions:

  1. How does GEAIRM use the KivaNura Agent Ontology without creating a second Agent taxonomy?
  2. How may an Agent participate in an AI Execution without becoming synonymous with Artificial Intelligence, AI Primitive or AI Product?
  3. How does AIPS AgencyAutonomy differ from Agent Ontology AutonomyLevel and ControlMode?
  4. How should the lexical collisions around AutonomyLevel and ControlMode in AIPS and the Agent Ontology be handled?
  5. What is a Tool in the enterprise-AI reference model, and is Tool an intrinsic type or a contextual role?
  6. How do Tool use, delegated authority and human/system oversight relate without implying one another?
  7. Which authority semantics can M08 establish without pre-empting M10 Decision/Action semantics or M13 governance semantics?

M08 deliberately does not define an Agent runtime architecture, Tool registry, orchestration plane, authorization service or Agent platform. Those are architecture or implementation concerns reserved for GEAIRA and concrete implementations.


151. Source Semantics and Authority Resolution​

M08 is governed by the M01 authority boundary.

The KivaNura Agent Ontology remains authoritative for Agent identity and Agent taxonomy. Its current 0.1 ontology establishes, among other concepts:

  • Agent;
  • IndividualAgent and CollectiveAgent;
  • LivingAgent, ArtificialAgent, HybridAgent and SimulatedAgent;
  • SoftwareAgent and RoboticAgent;
  • ModelTriggeredAgent;
  • StaticOrchestratorAgent;
  • DynamicReasoningAgent;
  • AutonomousCognitiveAgent;
  • AutonomyLevel with Assisted, SemiAutonomous, Autonomous;
  • ControlMode with HumanInLoop, HumanOnLoop, FullyAutomated;
  • ToolUsePattern with FixedTools, DynamicToolUse;
  • IntelligenceStyle, SubstrateType and LegalStatus facets.

AIPS separately establishes Agency / Autonomy (AR) as an AI Product Version classification whose values are:

  • Inform;
  • Recommend;
  • Assist;
  • ActWithApproval;
  • AutonomousAction.

AIPS describes AR as the authority exercised by the AI Product Version and treats it as an orthogonal Product-governance dimension rather than AI Product Type.

The AIPS core also retains classes named AutonomyLevel, ControlMode and AuthorityScope for detailed behavioural/governance semantics. These AIPS terms occupy the AIPS namespace and are not the same semantic identities as the similarly named Agent Ontology terms. The current AIPS JSON schema marks hasAutonomyLevel and hasControlMode as deprecated compatibility fields while the higher-level AR dimension is the normalized Product ontology classification.

M08 therefore adopts this authority treatment:

ConcernSemantic authorityGEAIRM treatment
Agent identity and taxonomyKivaNura Agent OntologyREUSED
Agent Autonomy LevelKivaNura Agent OntologyREUSED
Agent Control ModeKivaNura Agent OntologyREUSED
Agent Tool Use PatternKivaNura Agent OntologyREUSED
AI Product Version Agency / AutonomyAIPSREUSED
AIPS detailed AutonomyLevel / ControlMode compatibility semanticsAIPSREFERENCED only where needed; no equivalence to Agent Ontology terms
Generic Tool semanticsno sufficient canonical entity in the examined baselineresolved by M08 as a GEAIRM contextual Role
Generic delegated authority needed for Agent participationno single sufficient generic concept in the examined baselineresolved by M08 at minimum reference-model level
Detailed decision/action authorization semanticsfuture M10RESERVED
Governance policy/control implicationsfuture M13RESERVED

M08-R01 - Agent authority preservation: GEAIRM MUST preserve the KivaNura Agent Ontology as the semantic authority for Agent identity, Agent taxonomy and its canonical Agent facets.

M08-R02 - AIPS AR authority preservation: GEAIRM MUST preserve AIPS authority for AgencyAutonomy and its Product-Version classification values.


152. Agent​

152.1 Authoritative meaning​

GEAIRM reuses the KivaNura Agent Ontology Agent concept rather than minting a GEAIRM Agent class.

The authoritative ontology defines an Agent as an entity capable of perceiving, deciding and acting toward goals. It further aligns Agent with PROV, FOAF and Schema.org Agent concepts through its ontology hierarchy.

GEAIRM uses Agent as a first-class participant in enterprise AI while preserving the external identity.

152.2 Element record​

FieldValue
Preferred labelAgent
Element kindConcept
DefinitionKivaNura Agent Ontology authoritative definition applies.
Semantic authorityKivaNura Agent Ontology
Authority identifierAgent Ontology Agent identity
Representation dispositionREUSED
Normative statusreferenced/canonical through Agent Ontology
Source sliceAgent Ontology; integrated by GEAIRM M08
Constraints / invariantsAgent is not synonymous with Artificial Intelligence, AI Primitive, AI Product, AgenticGoalDriven Execution Mode or autonomous action authority.

152.3 Agent identity is not AI identity​

The Agent Ontology is intentionally broader than Artificial Intelligence. It includes Living Agents, Artificial Agents, Hybrid Agents and Collective Agents.

Therefore:

Agent != Artificial Intelligence
Agent != AI Primitive
Agent != AI Product

An Agent may participate in Artificial Intelligence, may depend on AI Primitives, may itself separately satisfy another authoritative classification, or may operate without qualifying AI semantics. None of those identities follows merely from Agent status.

Likewise, ArtificialAgent describes the substrate/implementation nature of an Agent. It MUST NOT be treated as an automatic assertion that the Agent is itself an AIPS AI Primitive or AI Product.

M08-R03 - Agent/AI non-collapse: Agent identity, including Artificial Agent identity, MUST NOT by itself establish Artificial Intelligence, AI Primitive or AI Product identity.


153. Agent Participation in AI Execution​

M06 introduced AI Execution as the context-bounded occurrence in which Artificial Intelligence is invoked, operated or enacted. M08 now introduces a generic relationship allowing an Agent to participate in such an occurrence without redefining Agent identity.

153.1 participatesInAIExecution​

FieldValue
Preferred labelparticipates in AI Execution
DefinitionRelates an Agent to an AI Execution in which that Agent materially participates in perceiving, initiating, coordinating, deciding, invoking, supervising, acting or otherwise contributing to the occurrence in the declared context.
Semantic authorityGEAIRM
Subject kindsAgent Ontology Agent
Object kindsGEAIRM AI Execution
Directionalitydirected: Agent -> AI Execution
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M08
Source sliceM08

The relationship is deliberately broad enough to model Human, Artificial, Hybrid and Collective Agent participation without introducing architecture-specific execution roles.

participatesInAIExecution does not imply that:

  • the Agent is the sole executor;
  • the Agent is an AI Primitive;
  • the Agent is an AI Product;
  • the Agent has autonomous authority;
  • the AI Execution is Agentic/Goal-driven;
  • the Agent caused every effect of the execution;
  • the Agent is authorized to perform every action involved;
  • the Agent owns the context, primitives or Products participating in the execution.

M08-R04 - Agent participation restraint: participatesInAIExecution MUST NOT be interpreted as sole execution, Product qualification, primitive identity, AgenticGoalDriven mode or delegated authority.


154. Agent Taxonomy Participation Without Semantic Reclassification​

GEAIRM may reference Agent Ontology subclasses where they are useful to enterprise-AI scenarios, but it does not reclassify them into AIPS axes merely from their names.

154.1 Model-Triggered Agent​

ModelTriggeredAgent is an Agent Ontology subtype describing a software Agent that invokes predefined models based on rules or triggers.

Its Agent subtype does not by itself imply:

  • LEARNED_MODEL AI Primitive Type;
  • any particular Intelligence Capability;
  • AgenticGoalDriven Execution Mode;
  • AI Product qualification.

A Model-Triggered Agent may invoke a model without the Agent itself being the model or AI Product.

154.2 Static Orchestrator Agent​

StaticOrchestratorAgent executes a predefined workflow or pipeline.

Static orchestration does not by itself establish Artificial Intelligence or AI Primitive status. This remains consistent with M04's distinction between automation and Artificial Intelligence.

154.3 Dynamic Reasoning Agent​

DynamicReasoningAgent is an Agent Ontology subtype whose authoritative definition includes planning, reasoning and dynamic action composition.

GEAIRM MUST NOT nevertheless infer AIPS Reasoning, Planning, a specific AI Primitive Type, or a particular Agency/Autonomy value solely from the subtype label. If an enterprise AI assertion requires those AIPS dimensions, the corresponding relationships/classifications SHOULD be stated explicitly or justified by an authoritative mapping.

154.4 Autonomous Cognitive Agent​

AutonomousCognitiveAgent is an Agent Ontology subtype whose definition includes memory, learning capability and adaptive behaviour over time.

The word Autonomous in the class name MUST NOT be used as evidence that:

  • AIPS AR = AutonomousAction;
  • Agent Ontology Autonomy Level = Autonomous for every instance without the ontology/profile asserting it;
  • the Agent has unrestricted action authority;
  • the Agent is an AI Product.

The authoritative Agent subtype meaning remains intact, but cross-ontology classification requires explicit semantics.

M08-R05 - Agent-subtype mapping restraint: Agent subtype names or definitions MUST NOT be converted automatically into AIPS Intelligence Capability, Execution Mode, Agency/Autonomy, AI Primitive Type or AI Product Type classifications.


155. Agency as an Editorial Umbrella, Not a Standalone GEAIRM Entity​

The word agency is overloaded across philosophy, organizational theory, legal usage, control systems and AI governance. AIPS also uses the combined dimension label Agency / Autonomy.

M08 therefore does not mint a standalone canonical GEAIRM entity named Agency.

Instead, GEAIRM models the operationally useful semantics through explicit concepts and classifications:

  • Agent identity;
  • Agent Autonomy Level;
  • Agent Control Mode;
  • AIPS Agency / Autonomy classification;
  • Delegated Authority;
  • Tool use;
  • participation in AI Execution;
  • later M10 decision/action authorization semantics.

Lowercase agency MAY appear in explanatory prose as an umbrella for an Agent's or AI participant's capacity to influence decisions or actions, but such prose MUST NOT imply a geairm:Agency class.

M08-R06 - No standalone Agency entity: GEAIRM 0.1 MUST NOT mint a generic Agency entity merely to unify semantically different autonomy, authority and Agent concepts.


156. AIPS Agency / Autonomy (AR)​

156.1 Authority and scope​

AIPS AgencyAutonomy is a Product-Version classification answering:

What authority does the AI Product Version exercise?

The current AIPS values are:

Inform
Recommend
Assist
ActWithApproval
AutonomousAction

M08 reuses this dimension but does not broaden the domain of the AIPS hasAgencyAutonomy property beyond AIProductVersion.

156.2 Element record​

FieldValue
Preferred labelAgency / Autonomy
Element kindClassification
DefinitionAIPS-controlled Product-Version dimension describing the authority exercised by the AI Product Version.
Semantic authorityAIPS
Authority identifieraips:AgencyAutonomy
Representation dispositionREUSED
Canonical valuesInform, Recommend, Assist, ActWithApproval, AutonomousAction
Constraints / invariantsAR is not Agent identity, Agent Autonomy Level, Agent Control Mode, Intelligence Effect, Execution Mode, Product Type or general delegated-authority identity.

156.3 AR is not Agent Autonomy Level​

AIPS AR and Agent Ontology Autonomy Level answer different questions:

AIPS Agency / Autonomy
What authority does this AI Product Version exercise?

Agent Ontology Autonomy Level
To what degree does this Agent operate independently?

An Agent may operate independently while the AI Product it uses is permitted only to Inform or Recommend. Conversely, a Product Version may be classified ActWithApproval while the Agent that participates in execution has a distinct Agent-level operational autonomy profile.

No universal mapping is established by M08.

M08-R07 - AR/autonomy separation: AIPS AgencyAutonomy and Agent Ontology AutonomyLevel MUST NOT be treated as equivalent classifications or deterministically mapped to one another.

M08-R08 - Product-property preservation: GEAIRM MUST NOT broaden the AIPS hasAgencyAutonomy property beyond its Product-Version semantic scope merely to characterize Agents.


157. Agent Autonomy Level​

The Agent Ontology AutonomyLevel is a facet describing the degree of independent operation of an Agent.

The current ontology values are:

  • Assisted;
  • SemiAutonomous;
  • Autonomous.

The authoritative Agent Ontology property is hasAutonomyLevel with domain Agent and range AutonomyLevel.

157.1 Autonomy is not authority scope​

Operational independence and delegated authority are related in real systems but are not the same concept.

An Agent may be operationally autonomous within a very narrow delegated authority scope. Another Agent may require human participation for routine operation while still being permitted to perform a consequential action once that human approval is present.

Therefore:

Agent Autonomy Level != Delegated Authority
Agent Autonomy Level != AIPS Agency / Autonomy
Agent Autonomy Level != Agent Control Mode

157.2 Autonomy is not Action Effect​

An Agent classified Autonomous does not necessarily produce DigitalAction or PhysicalAction effects in every execution. It may autonomously gather information, reason, generate content or produce recommendations while remaining unable to act externally.

M08-R09 - Autonomy/authority distinction: Agent Autonomy Level MUST NOT by itself establish the scope of decisions or actions the Agent is permitted to perform.

M08-R10 - Autonomy/effect distinction: Agent Autonomy Level MUST NOT deterministically imply any AIPS Intelligence Effect.


158. Agent Control Mode​

The Agent Ontology ControlMode facet describes the oversight and control mechanism governing an Agent.

The current ontology values are:

  • HumanInLoop;
  • HumanOnLoop;
  • FullyAutomated.

The authoritative Agent Ontology property is hasControlMode with domain Agent and range ControlMode.

158.1 Control Mode is not Intelligence Capability Control​

AIPS Control is an Intelligence Capability describing a kind of intelligence provided.

Agent Ontology ControlMode describes how an Agent is overseen or controlled.

Governance Control can also refer to policy/control mechanisms.

These remain distinct:

Agent Control Mode
!= AIPS Control Intelligence Capability
!= governance Control

158.2 Human-in-the-Loop is not automatically ActWithApproval​

HumanInLoop and AIPS ActWithApproval often appear together in implementations, but they answer different questions.

  • HumanInLoop characterizes the Agent's governing control/oversight mode.
  • ActWithApproval characterizes the AI Product Version's Agency / Autonomy level.

M08 does not assert equivalence.

158.3 Fully Automated is not Autonomous Action authority​

An Agent may operate without routine human oversight and still be limited to monitoring, scoring, generation or recommendation. Therefore FullyAutomated does not imply AIPS AutonomousAction.

M08-R11 - Control-mode separation: Agent Control Mode MUST remain distinct from AIPS Intelligence Capability Control, AIPS Agency/Autonomy and governance-control semantics.

M08-R12 - HITL/approval non-equivalence: HumanInLoop MUST NOT be treated as semantically equivalent to AIPS ActWithApproval.

M08-R13 - Automation/authority non-equivalence: FullyAutomated MUST NOT by itself establish AutonomousAction authority.


159. AIPS and Agent-Ontology Autonomy/Control Name Collision​

The current AIPS core also contains classes named aips:AutonomyLevel and aips:ControlMode, while the KivaNura Agent Ontology independently defines its own AutonomyLevel and ControlMode under a different namespace.

These are different semantic identities.

The AIPS specification uses its detailed behavioural fields for Product-focused behavioural/governance description, while the current AIPS schema marks hasAutonomyLevel and hasControlMode as deprecated compatibility fields in favour of normalized APT/IC/AR/MO/EM semantics.

GEAIRM therefore applies the following rule:

agent:AutonomyLevel != aips:AutonomyLevel
agent:ControlMode != aips:ControlMode

where != means no semantic equivalence is established by GEAIRM.

GEAIRM uses the Agent Ontology terms for generic Agent autonomy and control semantics. It references AIPS terms only where a Product-specific AIPS record requires them. M08 does not create a crosswalk between the two sets.

M08-R14 - Namespace collision discipline: Lexically identical AutonomyLevel or ControlMode terms from AIPS and the Agent Ontology MUST NOT be treated as the same semantic identity.

M08-R15 - No implicit crosswalk: GEAIRM MUST NOT infer mappings between AIPS detailed autonomy/control fields and Agent Ontology facets solely from shared labels.


160. Delegated Authority​

M01 identified generic Authority as a semantic-gap candidate because AIPS contains Authority Scope, Action Authority risk drivers and Agency/Autonomy semantics while Agent semantics contain control and autonomy facets. M08 resolves only the minimum concept needed to model Agent participation without pre-empting M10 decision/action semantics.

160.1 Canonical definition​

Delegated Authority is the bounded permission or mandate under which an Agent is permitted to exercise specified decision, invocation or action scope within a declared context, subject to applicable constraints and oversight.

Delegated Authority is deliberately narrower than a universal legal or organizational theory of authority. It models the enterprise-AI fact that independent operation and action permission are not the same thing.

160.2 Element record​

FieldValue
Preferred labelDelegated Authority
Element kindConcept
DefinitionThe bounded permission or mandate under which an Agent is permitted to exercise specified decision, invocation or action scope within a declared context, subject to applicable constraints and oversight.
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM08
Constraints / invariantsDelegated Authority is not Agent Autonomy Level, Control Mode, AIPS AR, Instruction, Tool Use Pattern, Product status or unrestricted authorization.

160.3 Scope discipline​

Delegated Authority MUST be interpreted as context-bounded. It may be constrained by task, time, resource, action kind, environment, organizational scope or other applicable context.

M08 does not yet define the detailed semantics of decision authority, action authorization, approval, revocation or policy enforcement. Those are elaborated in M10 and M13.

M08-R16 - Authority/autonomy separation: Delegated Authority MUST remain distinct from Agent Autonomy Level and AIPS Agency/Autonomy classification.

M08-R17 - Authority/context discipline: Delegated Authority MUST be interpreted within a declared or otherwise authoritative scope; it MUST NOT be presumed globally unrestricted.


161. operatesUnderDelegatedAuthority​

M08 introduces a relationship connecting Agent participation to the bounded authority under which it operates.

FieldValue
Preferred labeloperates under Delegated Authority
DefinitionRelates an Agent to a Delegated Authority that bounds the Agent's permitted decision, invocation or action scope in the declared context.
Semantic authorityGEAIRM
Subject kindsAgent Ontology Agent
Object kindsGEAIRM Delegated Authority
Directionalitydirected: Agent -> Delegated Authority
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M08
Source sliceM08

An Agent may operate under different Delegated Authorities in different contexts. M08 therefore establishes no universal single-valued cardinality.

The relationship does not imply:

  • that the Agent uses every permission in the authority scope;
  • that every contemplated action is automatically authorized;
  • that the authority cannot be conditional, revocable or time-bounded;
  • that the Agent is autonomous;
  • that the Agent is an AI Product;
  • that the Agent's Control Mode can be inferred.

M08-R18 - Delegated-authority relationship restraint: operatesUnderDelegatedAuthority MUST NOT imply unrestricted action permission, autonomous operation, Product status or a particular Control Mode.


162. Tool as a Contextual Role​

M01 identified Tool as a semantic-gap candidate because AIPS uses tools in dependency and constitution semantics, and the Agent Ontology defines Tool Use Pattern, but the examined baseline does not define one generic Tool entity class.

M08 resolves this gap by modelling Tool as a Role, not as an intrinsic superclass.

162.1 Canonical definition​

Tool is a contextual Role played by a resource, capability, service, Product, primitive or other invocable participant when it is made available for use by an Agent or AI Execution to obtain information, perform computation, invoke a capability, interact with an external system, or cause a permitted effect in a declared context.

This treatment is intentionally role-based because the same underlying thing may be:

  • a Product in one context;
  • an AI Primitive under AIPS semantics;
  • a service or API;
  • a data or information resource;
  • a Tool when used by an Agent or AI Execution.

Playing the Tool Role does not alter intrinsic identity.

162.2 Element record​

FieldValue
Preferred labelTool
Element kindRole
DefinitionA contextual Role played by an invocable resource or capability made available for use by an Agent or AI Execution in a declared context.
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM08
Constraints / invariantsTool Role is not Agent identity, AI Primitive Type, Product Family, authorization, action authority or architecture component type.

162.3 Tool Role does not imply AI​

A calculator, database API, payment API, search service, AI Product, model-serving endpoint or physical actuator could play a Tool Role depending on the scenario.

Therefore:

Tool Role != Artificial Intelligence
Tool Role != AI Primitive
Tool Role != AI Product

The resource playing the Tool Role retains its own semantic authority and identity.

M08-R19 - Tool role non-intrinsicality: Tool MUST be treated as a contextual Role in GEAIRM 0.1; playing the Tool Role MUST NOT reclassify the underlying resource's intrinsic type or Product Family.

M08-R20 - Tool/AI non-collapse: Tool Role MUST NOT by itself establish Artificial Intelligence, AI Primitive or AI Product status.


163. usesTool​

M08 introduces a generic semantic relationship for actual Tool use at reference-model level.

FieldValue
Preferred labeluses Tool
DefinitionRelates an Agent or AI Execution to a resource playing the Tool Role that is invoked or used in the declared context.
Semantic authorityGEAIRM
Subject kindsAgent Ontology Agent or GEAIRM AI Execution
Object kindsa resource playing the GEAIRM Tool Role
Directionalitydirected: Agent/AI Execution -> Tool-role resource
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M08
Source sliceM08

usesTool describes use, not entitlement or authority.

It does not imply that:

  • the Tool is owned by the Agent;
  • the Tool is authorized for every possible action;
  • the Tool is an AI Primitive;
  • the Tool is a Product;
  • the Tool is dynamically selected;
  • the Agent has autonomous authority;
  • the Tool's outputs are correct or trusted;
  • the Tool invocation itself constitutes an Intelligence Capability.

M08-R21 - Tool-use authority restraint: usesTool MUST NOT be interpreted as evidence of entitlement, authorization, delegated authority or unrestricted permission to invoke the Tool.


164. Agent Tool Use Pattern​

The Agent Ontology ToolUsePattern facet describes how an Agent utilizes tools or external resources.

The machine-readable Agent Ontology 0.1 establishes the canonical values:

  • FixedTools;
  • DynamicToolUse.

The interactive Agent documentation also describes a human-readable None case, but the current 0.1 ontology does not mint a corresponding ToolUsePattern individual. M08 therefore does not create or infer a canonical None value under the Agent Ontology namespace.

164.1 Fixed Tools​

FixedTools describes an Agent whose Tool set/pattern is predefined or otherwise fixed according to the Agent Ontology semantics.

It does not imply that the Agent invokes every available Tool in every execution.

164.2 Dynamic Tool Use​

DynamicToolUse describes an Agent that can select, compose or discover Tools/external resources dynamically according to the Agent Ontology semantics.

Dynamic Tool Use does not imply:

  • unrestricted Tool discovery;
  • permission to invoke any discovered Tool;
  • autonomous action authority;
  • AIPS AutonomousAction;
  • a particular AI Primitive Type;
  • a Tool registry architecture.

M08-R22 - Tool-use-pattern authority: GEAIRM MUST preserve Agent Ontology authority for Tool Use Pattern and its machine-readable canonical values.

M08-R23 - Tool-use-pattern/authority separation: Fixed or Dynamic Tool Use MUST NOT by itself establish delegated authority, entitlement or action authorization.

M08-R24 - No ungoverned ToolUsePattern extension: GEAIRM MUST NOT mint missing Agent-Ontology Tool Use Pattern values under the Agent namespace merely because explanatory documentation mentions additional cases.


165. Tool Use, Instructions and Authority Are Separate​

M07 established that an Instruction does not by itself grant authority. M08 extends that principle to Tool use.

The following are distinct semantic questions:

Instruction
What behaviour is requested or directed?

Tool Role / Tool Use
What resource is used to perform or support the behaviour?

Delegated Authority
What decision/invocation/action scope is permitted?

Agent Control Mode
What oversight/control mechanism governs the Agent?

Agent Autonomy Level
How independently does the Agent operate?

A Prompt may instruct an Agent to invoke a Tool, but neither the Prompt nor the Tool's availability proves that the invocation is authorized.

Likewise, possession of Delegated Authority does not imply that a particular Tool is available, fit or selected.

M08-R25 - Instruction/tool/authority separation: Instruction, Tool use and Delegated Authority MUST remain distinct; none MUST be inferred solely from the presence of another.


166. Agency / Autonomy, Autonomy Level, Control Mode and Delegated Authority Are Orthogonal​

M08 establishes four distinct behavioural/authority axes:

AxisAuthorityQuestion answered
AIPS Agency / Autonomy (AR)AIPSWhat authority does the AI Product Version exercise?
Agent Autonomy LevelAgent OntologyHow independently does the Agent operate?
Agent Control ModeAgent OntologyWhat oversight/control mechanism governs the Agent?
Delegated AuthorityGEAIRMWithin what bounded decision/invocation/action scope is the Agent permitted to operate?

These axes may correlate in specific implementations but are not reducible to one another.

Illustrative combinations include:

Agent Autonomy Level: Autonomous
Agent Control Mode: HumanOnLoop
AIPS AR: Recommend

Meaning:
the Agent may operate independently under supervision,
while the Product Version's authority remains recommendation-only.
Agent Autonomy Level: SemiAutonomous
Agent Control Mode: HumanInLoop
AIPS AR: ActWithApproval

Meaning:
the Agent operates independently only within bounded steps,
and material action requires approval.
Agent Control Mode: FullyAutomated
AIPS AR: Inform

Meaning:
the Agent may run without routine human oversight,
while the Product Version remains limited to informing rather than acting.

These are illustrative combinations, not mandatory mappings.

M08-R26 - Orthogonality: AIPS AR, Agent Autonomy Level, Agent Control Mode and Delegated Authority MUST remain independently representable and MUST NOT be collapsed into one ordinal scale.


167. Agentic/Goal-Driven Execution Mode Is Not Agent Identity​

M06 already established that AgenticGoalDriven is an AIPS Execution Mode and does not establish Agent identity.

M08 strengthens the reciprocal distinction:

  • an AI Execution may be AgenticGoalDriven without GEAIRM inferring a particular Agent subtype;
  • an Agent may participate in an AI Execution whose mode is Interactive, Continuous, Embedded or another valid mode rather than Agentic/Goal-driven;
  • Agent identity does not force a particular Execution Mode.

Therefore:

Agent != AgenticGoalDriven
AgenticGoalDriven != Agent

The first term is an entity identity; the second is an execution classification.

M08-R27 - Agent/mode separation: Agent identity and AgenticGoalDriven Execution Mode MUST remain semantically distinct and MUST NOT be inferred from one another.


168. Agent, AI Primitive and Product May Overlap Only Through Independent Qualification​

A real-world resource may participate under more than one semantic identity where the applicable authorities independently justify each identity.

For example, an artificial software resource might:

  • qualify as an Agent under the Agent Ontology;
  • depend on or incorporate one or more AI Primitives under AIPS;
  • itself separately satisfy AI Primitive semantics;
  • be packaged as a Product under BPS;
  • qualify as an AI Product only if AIPS qualification is satisfied.

The overlaps must be explicitly justified rather than inherited through composition or labels.

Agent status
does not imply AI Primitive status

AI Primitive status
does not imply Agent status

Agent + Product
does not imply AI Product

Agent + AI Primitive
does not imply AI Product

Detailed dependency/composition modelling remains reserved for M11.

M08-R28 - Independent qualification: Where the same resource is asserted to be Agent, AI Primitive and/or Product, each semantic identity MUST be independently supported by its authoritative criteria.


169. Collective Agents and Multi-Agent Systems​

The Agent Ontology defines CollectiveAgent as an Agent composed of multiple entities coordinating actions toward shared goals, with specialized Human, Non-Human and Hybrid Collective Agent classes.

M08 preserves that ontology concept but does not equate every collection of interacting Agents with a Collective Agent.

A multi-Agent system or interaction graph may contain several Agents without necessarily being asserted as one CollectiveAgent identity.

Similarly:

Collective Agent
!= multi-Agent topology automatically
!= Compound Product
!= Composite Derived AI Product
!= EnterpriseOrchestrated topology scope

M11 will address dependency, composition and topology semantics in detail.

M08-R29 - Collective/multi-agent distinction: Multiple Agents participating in one scenario MUST NOT automatically be classified as one Collective Agent.

M08-R30 - Collective/Product distinction: Collective Agent identity MUST NOT imply Compound Product, Composite derivation or AI Product identity.


170. Agent Memory, Planning and Tool Use Do Not Determine Agent Identity​

M07 introduced generic Memory semantics. The Agent documentation also describes Agent memory facets in explanatory material, while the machine-readable Agent Ontology 0.1 currently defines only a subset of those broader documentation facets.

GEAIRM therefore distinguishes:

  • the GEAIRM Memory concept established in M07;
  • Agent Ontology canonical machine-readable Agent classes/facets;
  • explanatory Agent-profile documentation that may describe memory/planning characteristics.

Use of Memory, Planning or Tools does not independently create Agent identity. Agent identity remains governed by the Agent Ontology definition and taxonomy.

Likewise, an Agent can use Memory or Tools without those resources becoming part of the Agent's intrinsic class identity.

M08-R31 - Feature/Agent identity restraint: Memory use, planning, Tool use, dynamic selection or orchestration MUST NOT by themselves establish Agent identity or Agent subtype unless the Agent authority establishes the classification.


171. Agent Oversight Does Not Transfer Responsibility or Authority Automatically​

Control Mode describes oversight mechanics. It does not, by itself, answer who owns business accountability, legal responsibility, Product ownership or decision authority.

For example:

  • HumanOnLoop means human supervision/intervention is part of the Agent control mode;
  • it does not by itself identify the accountable legal person;
  • HumanInLoop does not mean the human automatically becomes the Product owner;
  • FullyAutomated does not remove enterprise accountability.

GEAIRM does not model full organizational accountability or legal responsibility in M08. Those concerns may be linked through governance models where applicable.

M08-R32 - Oversight/accountability separation: Agent Control Mode MUST NOT be used as a shortcut for organizational accountability, legal responsibility, Product ownership or decision authority.


172. Canonical M08 Reference View​

The M08 reference view is:

                       KivaNura Agent Ontology
+---------------------------+
| Agent |
| external authority |
+-------------+-------------+
|
participatesInAIExecution
|
v
+----------------+
| AI Execution |
| GEAIRM-owned |
+-------+--------+
|
+--------------+--------------+
| |
usesTool M06/M07 relations
| |
v v
Resource playing Tool Role Primitive / Capability /
GEAIRM-owned Role Mode / Effect / Context

Agent facets:

agent:AutonomyLevel
Assisted | SemiAutonomous | Autonomous

agent:ControlMode
HumanInLoop | HumanOnLoop | FullyAutomated

agent:ToolUsePattern
FixedTools | DynamicToolUse

Agent
-- operatesUnderDelegatedAuthority --> Delegated Authority

AIPS Product-Version axis remains separate:

aips:AgencyAutonomy
Inform | Recommend | Assist | ActWithApproval | AutonomousAction

No arrows in this view establish equivalence among AR, Autonomy Level, Control Mode, Tool Use Pattern or Delegated Authority.


173. Illustrative Semantic Examples​

173.1 Autonomous Agent producing recommendations only​

Agent A
hasAutonomyLevel Autonomous [Agent Ontology]
hasControlMode HumanOnLoop [Agent Ontology]
participatesInAIExecution E1 [GEAIRM]

AI Product Version P1
hasAgencyAutonomy Recommend [AIPS]

The Agent may operate independently while the Product Version remains recommendation-only. No autonomous action authority follows.

173.2 Human-in-the-loop action with approval​

Agent A2
hasAutonomyLevel SemiAutonomous [Agent Ontology]
hasControlMode HumanInLoop [Agent Ontology]
operatesUnderDelegatedAuthority DA-1 [GEAIRM]

AI Product Version P2
hasAgencyAutonomy ActWithApproval [AIPS]

The two classifications are compatible but are not declared equivalent.

173.3 Dynamic Tool use under bounded authority​

Agent A3
hasToolUsePattern DynamicToolUse [Agent Ontology]
operatesUnderDelegatedAuthority DA-2 [GEAIRM]
usesTool Service-X [GEAIRM]

DynamicToolUse does not mean Service-X was authorized merely because it was discoverable. DA-2 must independently bound permitted use.

173.4 Static orchestrator without qualifying AI​

StaticOrchestratorAgent O1                 [Agent Ontology]
participatesInAIExecution? no assertion
AI Primitive? no assertion

The Agent subtype does not automatically establish Artificial Intelligence or AI Primitive status.

173.5 Tool that is also an AI Product​

FraudScoringProduct P
qualifies as AI Product [AIPS]

Within Agent execution E4:
P plays Tool Role [GEAIRM]
Agent A4 usesTool P [GEAIRM]

Tool Role does not replace the Product's intrinsic AI Product identity; it is a contextual role.

173.6 Agentic execution without Agent identity inference​

AI Execution E5
operatesInExecutionMode AgenticGoalDriven [GEAIRM/AIPS mode]

No particular Agent or Agent subtype is inferred unless an Agent participation assertion is separately established.


174. M08 Canonical Element Inventory​

ElementM03 kindAuthorityDispositionM08 status
AgentConceptKivaNura Agent OntologyREUSEDcanonical external
Individual AgentConcept/typeKivaNura Agent OntologyREUSEDcanonical external
Collective AgentConcept/typeKivaNura Agent OntologyREUSEDcanonical external
Living AgentConcept/typeKivaNura Agent OntologyREUSEDcanonical external
Artificial AgentConcept/typeKivaNura Agent OntologyREUSEDcanonical external
Hybrid AgentConcept/typeKivaNura Agent OntologyREUSEDcanonical external
Software AgentConcept/typeKivaNura Agent OntologyREUSEDcanonical external
Robotic AgentConcept/typeKivaNura Agent OntologyREUSEDcanonical external
Model-Triggered AgentConcept/typeKivaNura Agent OntologyREUSEDcanonical external
Static Orchestrator AgentConcept/typeKivaNura Agent OntologyREUSEDcanonical external
Dynamic Reasoning AgentConcept/typeKivaNura Agent OntologyREUSEDcanonical external
Autonomous Cognitive AgentConcept/typeKivaNura Agent OntologyREUSEDcanonical external
Agent Autonomy LevelClassificationKivaNura Agent OntologyREUSEDcanonical external
AssistedClassification valueKivaNura Agent OntologyREUSEDcanonical external
SemiAutonomousClassification valueKivaNura Agent OntologyREUSEDcanonical external
AutonomousClassification valueKivaNura Agent OntologyREUSEDcanonical external
Agent Control ModeClassificationKivaNura Agent OntologyREUSEDcanonical external
HumanInLoopClassification valueKivaNura Agent OntologyREUSEDcanonical external
HumanOnLoopClassification valueKivaNura Agent OntologyREUSEDcanonical external
FullyAutomatedClassification valueKivaNura Agent OntologyREUSEDcanonical external
Tool Use PatternClassificationKivaNura Agent OntologyREUSEDcanonical external
FixedToolsClassification valueKivaNura Agent OntologyREUSEDcanonical external
DynamicToolUseClassification valueKivaNura Agent OntologyREUSEDcanonical external
AIPS Agency / AutonomyClassificationAIPSREUSEDcanonical external
InformClassification valueAIPSREUSEDcanonical external
RecommendClassification valueAIPSREUSEDcanonical external
AssistClassification valueAIPSREUSEDcanonical external
ActWithApprovalClassification valueAIPSREUSEDcanonical external
AutonomousActionClassification valueAIPSREUSEDcanonical external
Agencyeditorial umbrellaGEAIRM editorial usagenot canonicalresolved - no standalone entity
Delegated AuthorityConceptGEAIRMOWNEDcanonical
ToolRoleGEAIRMOWNEDcanonical
participatesInAIExecutionRelationship TypeGEAIRMOWNEDcanonical
operatesUnderDelegatedAuthorityRelationship TypeGEAIRMOWNEDcanonical
usesToolRelationship TypeGEAIRMOWNEDcanonical

No executable GEAIRM identifiers are minted in M08. Stable machine identifiers remain deferred to the A1-A8 formalization programme after M17.


175. M08 Hard Distinctions​

The following distinctions are normative for subsequent slices:

Agent != Artificial Intelligence
Agent != AI Primitive
Agent != AI Product
Artificial Agent != AI Primitive automatically
Artificial Agent != AI Product automatically

Agent != AgenticGoalDriven Execution Mode
AgenticGoalDriven != Agent identity

AIPS Agency / Autonomy != Agent Autonomy Level
Agent Autonomy Level != Agent Control Mode
Agent Autonomy Level != Delegated Authority
Agent Control Mode != Delegated Authority

HumanInLoop != ActWithApproval
FullyAutomated != AutonomousAction
Autonomous Agent != AutonomousAction authority automatically

agent:AutonomyLevel != aips:AutonomyLevel
agent:ControlMode != aips:ControlMode

Agent Control Mode != AIPS Control Intelligence Capability
Agent Control Mode != governance Control

Tool Role != intrinsic resource type
Tool Role != AI Primitive
Tool Role != AI Product
Tool Use Pattern != Delegated Authority
DynamicToolUse != unrestricted tool permission

Instruction != Tool authorization
Instruction != Delegated Authority
Tool availability != Tool authorization

Collective Agent != arbitrary multi-Agent system
Collective Agent != Compound Product
Collective Agent != Composite Derived AI Product

Agent + Product != AI Product automatically
Agent + AI Primitive != AI Product automatically

As in previous slices, != denotes semantic non-equivalence rather than formal OWL disjointness unless later executable formalization explicitly authorizes stronger axioms.


176. M08 Normative Rules​

RuleRequirement
M08-R01GEAIRM MUST preserve the KivaNura Agent Ontology as semantic authority for Agent identity, taxonomy and canonical Agent facets.
M08-R02GEAIRM MUST preserve AIPS authority for AgencyAutonomy and its Product-Version classification values.
M08-R03Agent identity, including Artificial Agent identity, MUST NOT by itself establish Artificial Intelligence, AI Primitive or AI Product identity.
M08-R04participatesInAIExecution MUST NOT imply sole execution, Product qualification, primitive identity, AgenticGoalDriven mode or delegated authority.
M08-R05Agent subtype names or definitions MUST NOT automatically determine AIPS Intelligence Capability, Execution Mode, Agency/Autonomy, AI Primitive Type or AI Product Type classifications.
M08-R06GEAIRM 0.1 MUST NOT mint a standalone generic Agency entity merely to unify autonomy, authority and Agent concepts.
M08-R07AIPS AgencyAutonomy and Agent Ontology AutonomyLevel MUST NOT be treated as equivalent or deterministically mapped.
M08-R08GEAIRM MUST NOT broaden AIPS hasAgencyAutonomy beyond its Product-Version scope merely to characterize Agents.
M08-R09Agent Autonomy Level MUST NOT by itself establish permitted decision or action scope.
M08-R10Agent Autonomy Level MUST NOT deterministically imply any AIPS Intelligence Effect.
M08-R11Agent Control Mode MUST remain distinct from AIPS Control Intelligence Capability, AIPS Agency/Autonomy and governance-control semantics.
M08-R12HumanInLoop MUST NOT be treated as semantically equivalent to AIPS ActWithApproval.
M08-R13FullyAutomated MUST NOT by itself establish AIPS AutonomousAction.
M08-R14Lexically identical AIPS and Agent-Ontology AutonomyLevel or ControlMode terms MUST NOT be treated as the same semantic identity.
M08-R15GEAIRM MUST NOT infer crosswalks between AIPS detailed autonomy/control fields and Agent Ontology facets solely from shared labels.
M08-R16Delegated Authority MUST remain distinct from Agent Autonomy Level and AIPS Agency/Autonomy.
M08-R17Delegated Authority MUST be interpreted within a declared or authoritative scope and MUST NOT be presumed globally unrestricted.
M08-R18operatesUnderDelegatedAuthority MUST NOT imply unrestricted action permission, autonomous operation, Product status or a particular Control Mode.
M08-R19Tool MUST be treated as a contextual Role in GEAIRM 0.1; Tool Role MUST NOT alter the underlying resource's intrinsic identity or Product Family.
M08-R20Tool Role MUST NOT by itself establish Artificial Intelligence, AI Primitive or AI Product status.
M08-R21usesTool MUST NOT be interpreted as evidence of entitlement, authorization, delegated authority or unrestricted permission.
M08-R22GEAIRM MUST preserve Agent Ontology authority for Tool Use Pattern and its machine-readable canonical values.
M08-R23Fixed or Dynamic Tool Use MUST NOT by itself establish delegated authority, entitlement or action authorization.
M08-R24GEAIRM MUST NOT mint Agent-Ontology Tool Use Pattern values merely because explanatory material mentions cases absent from the machine-readable ontology.
M08-R25Instruction, Tool use and Delegated Authority MUST remain distinct; none MUST be inferred solely from the presence of another.
M08-R26AIPS AR, Agent Autonomy Level, Agent Control Mode and Delegated Authority MUST remain independently representable and MUST NOT be collapsed into one ordinal scale.
M08-R27Agent identity and AgenticGoalDriven Execution Mode MUST remain semantically distinct and MUST NOT be inferred from one another.
M08-R28Agent, AI Primitive and Product identities applying to the same resource MUST each be independently supported by their authoritative criteria.
M08-R29Multiple Agents participating in one scenario MUST NOT automatically be classified as one Collective Agent.
M08-R30Collective Agent identity MUST NOT imply Compound Product, Composite derivation or AI Product identity.
M08-R31Memory use, planning, Tool use, dynamic selection or orchestration MUST NOT by themselves establish Agent identity or Agent subtype.
M08-R32Agent Control Mode MUST NOT be used as a shortcut for organizational accountability, legal responsibility, Product ownership or decision authority.
M08-R33Tool use, Agent autonomy or Agentic execution MUST NOT override M07's rule that Instruction does not grant authority.
M08-R34M08 MUST NOT prescribe Tool registries, Agent runtimes, orchestration platforms or authorization-service architecture; those concerns belong to GEAIRA or implementation specifications.
M08-R35M08 Delegated Authority semantics MUST NOT be interpreted as a complete legal, organizational or policy authority model; detailed Decision/Action authority remains reserved for M10 and governance realization for M13/GEAIRA.
M08-R36M08 semantic non-equivalence statements MUST NOT be strengthened into formal disjointness unless later executable formalization explicitly authorizes that inference.

177. M08 Exit Criteria​

GEAIRM-M08 is complete when:

  • Agent identity and taxonomy remain governed by the KivaNura Agent Ontology;
  • Agent participation in AI Execution can be represented without creating a GEAIRM Agent shadow class;
  • Artificial Agent is separated from automatic AI Primitive or AI Product identity;
  • Agent subtypes remain distinct from AIPS Product/capability/mode classifications;
  • Agency is resolved without unnecessary creation of a universal standalone entity;
  • AIPS Agency / Autonomy is explicitly reused and remains Product-Version scoped;
  • Agent Autonomy Level is explicitly reused and separated from AR and authority scope;
  • Agent Control Mode is explicitly reused and separated from Intelligence Capability Control, AR and governance Control;
  • AIPS and Agent-Ontology namespace collisions around AutonomyLevel/ControlMode are explicitly handled without inferred equivalence;
  • Delegated Authority is minimally defined as a context-bounded enterprise-AI concept without pre-empting M10/M13;
  • Agent-to-Delegated-Authority participation can be represented;
  • Tool is resolved as a contextual Role rather than an intrinsic superclass;
  • Tool use can be represented without implying entitlement or authority;
  • Agent Ontology Tool Use Pattern is preserved without silently adding machine values;
  • AgenticGoalDriven Execution Mode remains distinct from Agent identity;
  • Collective Agent remains distinct from arbitrary multi-Agent topology and Product composition;
  • Agent/AI Primitive/Product overlaps require independent qualification;
  • no Tool registry, Agent runtime, orchestration platform or authorization architecture is prescribed;
  • M01-M07 authority, abstraction and conservative-inference rules remain intact.

M08 conclusion: these criteria are satisfied by this slice. GEAIRM now has a stable Agent participation model that preserves Agent Ontology authority, keeps AIPS Product-level Agency/Autonomy separate from Agent operational autonomy and control, introduces minimal Delegated Authority semantics, and resolves Tool as a contextual role without turning tool use into authority or Product identity.


178. Forward Work​

The next slice is:

GEAIRM-M09 - Embodiment, Environment & Control-Loop Model

M09 will reconcile AIPS Embodiment with Agent Ontology physical/robotic Agent semantics, resolve the generic Environment semantic gap, and model sensing, perception, actuation, digital/physical action and closed-loop interaction without collapsing Embodiment, Agent substrate, Intelligence Capability, Intelligence Effect or autonomous authority.


179. GEAIRM-M09 - Embodiment, Environment & Control-Loop Model​

179.1 Purpose of this slice​

M08 established Agent participation, Agent autonomy/control semantics, Delegated Authority and Tool use. M09 now establishes how enterprise AI relates to physical and digital environments, while preserving the distinct meanings of AIPS Embodiment, Agent Ontology substrate/robotic semantics, Intelligence Capability, Intelligence Effect, autonomy and authority.

M09 answers eight questions:

  1. How does GEAIRM reuse the AIPS Embodiment classification without turning it into a generic environment taxonomy?
  2. How does AIPS Embodiment differ from Agent Ontology RoboticAgent and SubstrateType?
  3. What generic Environment concept is required by the enterprise-AI reference model?
  4. How are sensing and AIPS Perception kept distinct?
  5. How are actuation and AIPS PhysicalAction / DigitalAction kept distinct?
  6. How may an AI Execution interact with an Environment without implying embodiment, autonomy or Agent identity?
  7. How is closed-loop interaction represented without equating every feedback loop with autonomous action?
  8. Which physical-safety and governance implications remain reserved for M13, GEAIRA and AIPS governance rules?

M09 does not prescribe sensors, actuators, robotics middleware, digital twins, control buses, edge runtimes, safety controllers or physical deployment topology. Those are architecture and implementation concerns.


180. Source Semantics and Authority Resolution​

M09 is governed by the M01 semantic-authority boundary and the distinctions established in M04-M08.

AIPS already defines Embodiment (EM) as the Product-Version dimension answering:

How does the AI Product relate to the physical world?

The current AIPS controlled values are:

NonEmbodied
PhysicalWorldPerceptive
EmbodiedActuating
ClosedLoopAutonomous

AIPS explanatory semantics characterize them as:

  • Non-embodied - operates in the digital domain;
  • Physical-world Perceptive - senses or interprets the physical world;
  • Embodied / Actuating - controls physical systems or devices;
  • Closed-loop Autonomous - continuously perceives, reasons and acts in the physical world.

AIPS also defines PhysicalAction and DigitalAction as Intelligence Effect values and Perception and Control as Intelligence Capability values. These axes remain orthogonal.

The KivaNura Agent Ontology independently defines:

  • RoboticAgent as a physically embodied Artificial Agent equipped with sensors and actuators;
  • SubstrateType with Living, Artificial and Hybrid values;
  • hasSubstrateType as an Agent facet.

The current machine-readable Agent Ontology 0.1 does not establish a generic Embodiment facet or values named SoftwareEmbodied, PhysicalEmbodied or DistributedEmbodied, even though explanatory Agent documentation discusses those notions. M09 therefore does not mint them under the Agent Ontology namespace.

M09 resolves the M01 Environment semantic-gap candidate by introducing a minimal generic GEAIRM Environment concept and associated interaction semantics.

ConcernSemantic authorityM09 treatment
AIPS EmbodimentAIPSREUSED
NonEmbodied / PhysicalWorldPerceptive / EmbodiedActuating / ClosedLoopAutonomousAIPSREUSED
Perception / ControlAIPS Intelligence CapabilityREUSED
DigitalAction / PhysicalActionAIPS Intelligence EffectREUSED
RoboticAgentKivaNura Agent OntologyREUSED
Agent Substrate TypeKivaNura Agent OntologyREUSED
Generic EnvironmentGEAIRMOWNED
Environment KindGEAIRMOWNED
SensingGEAIRMOWNED
ActuationGEAIRMOWNED
Control LoopGEAIRMOWNED
Safety controls / emergency stop / safe state obligationsAIPS governance / M13 / GEAIRAREFERENCED or RESERVED

M09-R01 - Embodiment authority preservation: GEAIRM MUST preserve AIPS authority for Embodiment, hasEmbodiment and the current AIPS Embodiment controlled values.

M09-R02 - Agent authority preservation: GEAIRM MUST preserve the Agent Ontology as authority for RoboticAgent, SubstrateType and Agent substrate semantics.


181. AIPS Embodiment​

181.1 Embodiment as a reused Classification​

FieldValue
Preferred labelEmbodiment
Element kindClassification
DefinitionAIPS-controlled Product-Version dimension describing the relationship of the intelligence to the physical world.
Semantic authorityAIPS
Authority identifieraips:Embodiment
Representation dispositionREUSED
Normative statusreferenced/canonical through AIPS
Source sliceAIPS; integrated by GEAIRM M09
Canonical valuesNonEmbodied, PhysicalWorldPerceptive, EmbodiedActuating, ClosedLoopAutonomous

AIPS hasEmbodiment is scoped to AIProductVersion and is functionally constrained in the current AIPS model. M09 does not broaden that property to Agent, AI Execution, Environment or AI Primitive.

M09-R03 - Product-property preservation: GEAIRM MUST NOT broaden AIPS hasEmbodiment beyond its AI Product Version domain merely to characterize generic AI execution, Agents or environments.

181.2 Embodiment is not intrinsic Product Type​

Embodiment remains orthogonal to AI Product Type and Product Intrinsic Nature. A Primary or Derived AI Product may be non-embodied, perceptive, actuating or closed-loop depending on its own governed semantics.

Likewise, changing Embodiment does not automatically change BPS Product Family or APT unless the applicable Product identity invariants and AIPS qualification rules independently establish such a change.

M09-R04 - Embodiment/Product-type orthogonality: AIPS Embodiment MUST NOT be treated as AI Product Type, Product Family, AI Primitive Type or Agent subtype.


182. Embodiment Values and Their Boundaries​

182.1 NonEmbodied​

NonEmbodied characterizes an AI Product Version operating without a governed physical-world embodiment relationship under the AIPS Embodiment dimension.

It does not imply:

  • absence of digital actions;
  • absence of external systems;
  • absence of human users;
  • absence of sensors elsewhere in a wider system;
  • low risk;
  • absence of Agents.

A non-embodied Product may still cause consequential digital effects.

182.2 PhysicalWorldPerceptive​

PhysicalWorldPerceptive characterizes a Product Version whose intelligence senses or interprets the physical world according to AIPS semantics.

M09 preserves a distinction between sensing and the AIPS Perception Intelligence Capability. Physical-world perceptiveness may involve both, but they are not the same semantic dimension.

182.3 EmbodiedActuating​

EmbodiedActuating characterizes a Product Version that controls or actuates physical systems/devices under AIPS semantics.

It does not by itself imply:

  • autonomous action authority;
  • Agent Ontology Autonomous;
  • Agent Control Mode FullyAutomated;
  • AIPS AutonomousAction;
  • a closed feedback loop;
  • RoboticAgent identity.

182.4 ClosedLoopAutonomous​

ClosedLoopAutonomous is an AIPS Embodiment value combining a physical-world closed-loop relationship with autonomous operation at the governed Product classification level.

M09 does not decompose or redefine the AIPS value. However, GEAIRM separately models the generic semantic concept Control Loop so that closed-loop interaction can also be represented outside the narrower AIPS Product classification.

Therefore:

Control Loop
!= ClosedLoopAutonomous Embodiment

A generic control loop may include human approval, bounded automation, non-AI control or other arrangements. The AIPS value is a specific Product-Version classification and carries its own governance implications.

M09-R05 - Closed-loop/autonomy separation: The generic GEAIRM Control Loop concept MUST NOT be treated as equivalent to AIPS ClosedLoopAutonomous Embodiment.

M09-R06 - Embodiment/autonomy separation: EmbodiedActuating or PhysicalWorldPerceptive MUST NOT by itself determine AIPS Agency/Autonomy, Agent Autonomy Level, Agent Control Mode or Delegated Authority.


183. Agent Embodiment, Robotic Agent and Substrate Are Separate from AIPS EM​

AIPS Embodiment and Agent Ontology semantics answer different questions.

AIPS Embodiment
How does an AI Product Version relate to the physical world?

RoboticAgent
What kind of Agent is this under the Agent Ontology?

Agent SubstrateType
What underlying substrate characterizes the Agent?

These concepts may correlate in concrete systems but are not interchangeable.

A RoboticAgent is defined by the Agent Ontology as physically embodied and equipped with sensors and actuators. That Agent identity does not automatically establish the AIPS Embodiment classification of any AI Product Version used by, contained in or exposed through the robot.

Likewise, an AI Product Version classified EmbodiedActuating does not automatically become an Agent, RoboticAgent or ArtificialAgent.

The Agent Ontology Artificial Substrate Type is broader than robotics and must not be read as AIPS Artificial Intelligence or AIPS Embodiment.

M09-R07 - Robotic/Embodiment non-equivalence: RoboticAgent identity MUST NOT automatically determine AIPS Embodiment, and AIPS Embodiment MUST NOT automatically establish RoboticAgent identity.

M09-R08 - Substrate/AI distinction: Agent Ontology Artificial Substrate Type MUST NOT be interpreted as Artificial Intelligence, AI Primitive identity or a particular AIPS Embodiment value.

M09-R09 - No undocumented Agent facet minting: GEAIRM MUST NOT mint Agent-Ontology embodiment values under the Agent namespace solely because explanatory documentation describes embodiment notions absent from the current machine-readable Agent Ontology.


184. Environment​

M01 identified Environment as a semantic-gap candidate. M09 resolves it as a minimal generic GEAIRM Concept.

184.1 Canonical definition​

Environment is a bounded external or surrounding domain of state, resources, entities and conditions with which an Agent or AI Execution may interact, from which it may receive observations or inputs, or upon which it may produce effects.

Environment is deliberately generic. It may be physical, digital or mixed. It is not an architecture runtime, deployment platform, world model, context object or Product.

184.2 Element record​

FieldValue
Preferred labelEnvironment
Element kindConcept
DefinitionA bounded external or surrounding domain of state, resources, entities and conditions with which an Agent or AI Execution may interact, receive observations/inputs from, or produce effects upon.
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM09
Constraints / invariantsEnvironment is not AI Context, M03 Modelling Context, AI Product, Agent, Tool, runtime platform or world model.

184.3 Environment is not AI Context​

M07 AI Context is the bounded set of semantically relevant resources, state, conditions, instructions and information made available to or considered by an AI Execution.

Environment is the external/surrounding domain with which the execution or Agent may interact.

A portion or representation of Environment state may contribute to AI Context, but the two concepts are not identical:

Environment state / observation
may contribute to
AI Context

but

Environment != AI Context

M09-R10 - Environment/context separation: Environment MUST remain distinct from AI Context and M03 Modelling Context; representing Environment state in context MUST NOT collapse the concepts.


185. Environment Kind​

M09 introduces a lightweight Classification for the broad domain in which an Environment exists.

185.1 Canonical values​

PhysicalEnvironment
DigitalEnvironment
MixedEnvironment
  • PhysicalEnvironment - an environment whose relevant state and effects materially exist in the physical world.
  • DigitalEnvironment - an environment whose relevant state and effects are represented and enacted within digital systems or computational domains.
  • MixedEnvironment - an environment in which relevant physical and digital state/effects are materially coupled.

This classification describes the Environment, not the AI Product Version's AIPS Embodiment.

For example, an AI Product may operate NonEmbodied while interacting with a DigitalEnvironment. A Robotic Agent may interact with a MixedEnvironment while a particular AI Product Version it invokes has its own independently governed Embodiment declaration.

M09-R11 - Environment-kind/Embodiment separation: Environment Kind MUST NOT be treated as a synonym or replacement for AIPS Embodiment.

M09-R12 - Environment-kind non-inheritance: Interaction with a PhysicalEnvironment MUST NOT by itself make an Agent a RoboticAgent, an AI Product EmbodiedActuating, or an effect PhysicalAction.


186. occursInEnvironment​

M09 introduces a generic relationship for the environment in which an AI Execution materially occurs or interacts.

FieldValue
Preferred labeloccurs in Environment
DefinitionRelates an AI Execution to an Environment in which the execution materially receives state/inputs, interacts, or produces effects in the declared semantic context.
Semantic authorityGEAIRM
Subject kindsGEAIRM AI Execution
Object kindsGEAIRM Environment
Directionalitydirected: AI Execution -> Environment
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M09
Source sliceM09

The relationship does not imply that the execution is embodied, actuating, autonomous, AgenticGoalDriven or physically located inside every Environment it interacts with.

An AI Execution may interact with multiple Environments or with a remote Environment through Tools or interfaces.

M09-R13 - Environment relationship restraint: occursInEnvironment MUST NOT imply AIPS Embodiment, Agent identity, physical co-location, autonomous authority or any particular Intelligence Effect.


187. Sensing​

187.1 Canonical definition​

Sensing is the acquisition of signals, measurements, observations or state from an Environment through a sensor, interface or other observation mechanism for potential use by an Agent or AI Execution.

Sensing is an activity/process concept. It does not itself imply Artificial Intelligence.

A conventional temperature sensor, telemetry feed or API poll can participate in Sensing without being an AI Primitive.

187.2 Sensing is not Perception​

AIPS Perception is an Intelligence Capability. Sensing is the acquisition of signals/state that may provide input to a Perception capability.

Sensing
acquires observations/state

Perception
Intelligence Capability
interprets or derives intelligence from relevant input

A system may sense without applying qualifying AI Perception. Conversely, a Perception capability may operate on pre-recorded or previously acquired inputs rather than live sensing.

M09-R14 - Sensing/Perception separation: Sensing MUST remain distinct from the AIPS Perception Intelligence Capability and MUST NOT by itself establish Artificial Intelligence or AI Primitive status.


188. sensesEnvironment​

FieldValue
Preferred labelsenses Environment
DefinitionRelates an Agent or AI Execution to an Environment from which it acquires signals, observations or state through Sensing in the declared context.
Semantic authorityGEAIRM
Subject kindsAgent Ontology Agent or GEAIRM AI Execution
Object kindsGEAIRM Environment
Directionalitydirected: Agent/AI Execution -> Environment
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M09
Source sliceM09

sensesEnvironment does not imply that the subject exercises the Perception Intelligence Capability, that the observations enter AI Context, or that they are trusted/grounded.

Where sensed observations are made available to an AI Execution, M07 context relationships may separately represent their inclusion or grounding.

M09-R15 - Sensing/context restraint: Sensing an Environment MUST NOT automatically imply inclusion, grounding, correctness or causal use of every sensed observation in AI Context.


189. Actuation​

189.1 Canonical definition​

Actuation is the application of a command, control signal or physical intervention through an actuator or controlled mechanism in order to change or influence a physical system or Physical/Mixed Environment.

Actuation is an interaction/process concept. It is not itself an AIPS Intelligence Effect value and is not synonymous with authority to act.

189.2 Actuation vs Control capability​

AIPS Control is an Intelligence Capability. Actuation is the mechanism/process through which a physical change may be applied.

A Control capability may determine or contribute a control signal without directly actuating a device. Conversely, an actuator can execute a command generated by ordinary deterministic logic or human input.

Control capability != Actuation

189.3 Actuation vs PhysicalAction effect​

AIPS PhysicalAction answers what intelligence produces or causes at the Intelligence Effect dimension. Actuation describes the physical interaction mechanism/process through which change is applied.

A PhysicalAction may be realized through one or more actuations, but the concepts are not identical and M09 imposes no one-to-one mapping.

M09-R16 - Control/Actuation separation: AIPS Control Intelligence Capability MUST remain distinct from Actuation.

M09-R17 - Actuation/PhysicalAction separation: Actuation MUST remain distinct from AIPS PhysicalAction Intelligence Effect and MUST NOT be inferred solely from the presence of that effect classification.


190. actsUponEnvironment​

M09 introduces a generic relationship for an Agent or AI Execution materially affecting an Environment.

FieldValue
Preferred labelacts upon Environment
DefinitionRelates an Agent or AI Execution to an Environment that it materially changes, influences or attempts to influence through an action or actuation in the declared context.
Semantic authorityGEAIRM
Subject kindsAgent Ontology Agent or GEAIRM AI Execution
Object kindsGEAIRM Environment
Directionalitydirected: Agent/AI Execution -> Environment
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M09
Source sliceM09

The relationship may apply to Digital, Physical or Mixed Environments depending on the scenario. It therefore does not itself determine whether the effect is DigitalAction or PhysicalAction.

actsUponEnvironment also does not establish that the action was authorized, autonomous, successful or safe.

M09-R18 - Action/environment restraint: actsUponEnvironment MUST NOT by itself establish DigitalAction, PhysicalAction, AIPS Agency/Autonomy, Delegated Authority, success or safety.


191. Digital Action, Physical Action and Environment​

M06 integrated AIPS DigitalAction and PhysicalAction as Intelligence Effect values. M09 adds their environment boundary.

191.1 DigitalAction​

A DigitalAction is an Intelligence Effect whose material change is enacted in a digital/computational domain according to AIPS effect semantics.

DigitalAction does not imply:

  • NonEmbodied classification automatically;
  • Agent identity;
  • autonomous action authority;
  • Tool use;
  • absence of physical consequences downstream.

A digital action may indirectly lead to physical consequences through another system, but that causal chain does not retroactively make every upstream digital action a PhysicalAction.

191.2 PhysicalAction​

A PhysicalAction is an Intelligence Effect involving material action in the physical world according to AIPS effect semantics.

PhysicalAction does not by itself establish:

  • EmbodiedActuating or ClosedLoopAutonomous Embodiment;
  • RoboticAgent identity;
  • autonomous authority;
  • Control capability;
  • Agent Autonomy Level.

The applicable AIPS Product-Version Embodiment must remain explicitly governed through AIPS rather than inferred generically from one effect occurrence.

M09-R19 - Digital-action boundary: DigitalAction MUST NOT by itself determine AIPS Embodiment or Agent identity.

M09-R20 - Physical-action boundary: PhysicalAction MUST NOT by itself determine AIPS Embodiment, RoboticAgent identity, Control capability or autonomous authority.


192. Control Loop​

M09 introduces a generic semantic concept for repeated environment-interaction cycles.

192.1 Canonical definition​

Control Loop is a recurring interaction pattern in which Environment state is sensed or otherwise observed, relevant state is evaluated or interpreted, a response or control decision is formed, and an action may influence the Environment such that subsequent observations can reflect the resulting change.

The definition is intentionally broader than autonomous robotics. A Control Loop may be:

  • deterministic or AI-enabled;
  • digital, physical or mixed;
  • human-supervised or automated;
  • advisory or actuating;
  • open to intervention, approval or override.

Control Loop therefore does not itself establish Artificial Intelligence or autonomy.

192.2 Element record​

FieldValue
Preferred labelControl Loop
Element kindConcept / interaction pattern
DefinitionA recurring environment-interaction pattern in which observed state informs response formation and resulting action may influence later observed state.
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM09
Constraints / invariantsControl Loop is not AIPS ClosedLoopAutonomous, Agent Control Mode, Control Intelligence Capability, autonomous authority or architecture topology.

M09-R21 - Control-loop/AI distinction: Presence of a Control Loop MUST NOT by itself establish Artificial Intelligence, AI Primitive status or AI Product qualification.

M09-R22 - Control-loop/control-mode distinction: Control Loop MUST remain distinct from Agent Ontology Control Mode and AIPS Control Intelligence Capability.


193. participatesInControlLoop​

FieldValue
Preferred labelparticipates in Control Loop
DefinitionRelates an Agent or AI Execution to a Control Loop in which it materially participates in sensing, interpretation, response formation, supervision, decision or action in the declared context.
Semantic authorityGEAIRM
Subject kindsAgent Ontology Agent or GEAIRM AI Execution
Object kindsGEAIRM Control Loop
Directionalitydirected: Agent/AI Execution -> Control Loop
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M09
Source sliceM09

Participation in a Control Loop does not state which stage the participant performs and does not imply authority over the complete loop.

A Human Agent may participate as approver or supervisor; an AI Execution may participate only in perception or recommendation; another component may perform actuation.

M09-R23 - Loop-participation restraint: participatesInControlLoop MUST NOT imply control over every loop stage, autonomous authority, AgenticGoalDriven mode or closed-loop AIPS Embodiment.


194. Control Loop vs ClosedLoopAutonomous Embodiment​

The relationship between the generic GEAIRM Control Loop and AIPS ClosedLoopAutonomous is intentionally one-way only at the level of conceptual relevance, not formal equivalence.

An AI Product Version classified ClosedLoopAutonomous would ordinarily participate in a closed-loop physical-world interaction under its authoritative AIPS semantics. However, M09 does not modify AIPS to express that relationship as a new ontology axiom.

Conversely, a GEAIRM Control Loop may exist with:

  • HumanInLoop or HumanOnLoop supervision;
  • ActWithApproval authority;
  • non-AI deterministic control;
  • digital-only feedback;
  • recommendation-only AI participation;
  • multiple Agents or systems dividing loop responsibilities.

Therefore:

ClosedLoopAutonomous
implies stronger AIPS Product semantics

Control Loop
is a broader generic interaction concept

M09-R24 - No reverse inference: Participation in a Control Loop MUST NOT be used to infer ClosedLoopAutonomous Embodiment.


195. Perception, Sensing, Control and Action Form Distinct Axes​

M09 consolidates four related but distinct questions:

ConcernQuestionAuthority
SensingHow is Environment state acquired?GEAIRM
PerceptionWhat intelligence interprets or derives meaning from relevant input?AIPS Intelligence Capability
ControlWhat intelligence determines/control-oriented response?AIPS Intelligence Capability
ActuationHow is physical change applied?GEAIRM
DigitalAction / PhysicalActionWhat action effect is produced or caused?AIPS Intelligence Effect
EmbodimentHow does the AI Product Version relate to the physical world?AIPS
Agent Control ModeWhat oversight mechanism governs the Agent?Agent Ontology
Delegated AuthorityWhat bounded decision/invocation/action scope is permitted?GEAIRM

None of these axes universally determines another.

For example:

Sensing
may feed
Perception
may support
Control
may lead to
PhysicalAction
may be realized through
Actuation

is a valid illustrative chain, but not a mandatory universal pipeline.

M09-R25 - Interaction-axis orthogonality: Sensing, Perception, Control, Actuation, Intelligence Effect, Embodiment, Agent Control Mode and Delegated Authority MUST remain independently representable unless a narrower contextual assertion is explicitly established.


196. Physical-World Observation Can Contribute to AI Context​

M07 established that observations and resources may participate in AI Context. M09 connects Environment interaction without creating a duplicate observation ontology.

A sensed observation or Environment-state representation MAY:

  • participate as an AI Context Resource;
  • play a Grounding Source Role;
  • be retrieved later;
  • be retained in Memory.

These roles remain independent.

Sensing alone does not mean the resulting observation is trusted, grounded, authoritative or used in the AI Execution.

M09 therefore does not mint a universal Observation class in GEAIRM 0.1. Where another authoritative domain defines observation semantics, GEAIRM may reference that concept later.

M09-R26 - Observation minimality: GEAIRM 0.1 MUST NOT mint a universal Observation ontology merely to represent sensed Environment state where existing M07 context-resource semantics are sufficient.


197. Embodiment and Tool Use​

M08 established Tool as a contextual Role. A sensor, actuator, robotic subsystem, external service or control interface may play the Tool Role for an Agent or AI Execution.

Tool use and Embodiment remain separate:

usesTool Sensor-X
!= PhysicalWorldPerceptive automatically

usesTool Actuator-Y
!= EmbodiedActuating automatically

A Product Version's Embodiment is governed by AIPS based on the Product's actual physical-world relationship, not merely the presence of a Tool dependency.

Similarly, a Tool may itself be embodied or non-embodied under its own applicable semantics while serving another participant.

M09-R27 - Tool/Embodiment separation: Use of sensing or actuation Tools MUST NOT by itself determine the AIPS Embodiment of the Agent, AI Execution or AI Product Version using them.


198. Embodiment, Environment and Authority​

Physical-world interaction and authority are distinct concerns.

A Product or Agent may be technically capable of actuation without being authorized to perform a material action. Conversely, delegated authority may permit an action only under conditions, approval gates, safety constraints or bounded Environment scope.

Therefore:

physical capability != permission
actuation capability != authorization
physical action != autonomous authority
closed-loop participation != unrestricted autonomy

M08 Delegated Authority remains the generic bounded-permission concept. Detailed authorization, approval, Decision and Action semantics are resolved in M10. Governance controls, safety evidence, emergency intervention and obligation profiles remain under AIPS/M13 and future GEAIRA realization.

M09-R28 - Physical-capability/authority separation: Sensing, actuation, physical embodiment or control-loop participation MUST NOT by themselves establish permission or authorization to decide or act.


199. AIPS Physical-AI Governance Semantics Remain External​

The current AIPS baseline treats certain physical/autonomous combinations as material governance drivers. For example, AIPS validation requires higher risk treatment for combinations involving AutonomousAction + PhysicalAction and ClosedLoopAutonomous + PhysicalAction.

M09 does not duplicate those risk rules inside GEAIRM.

GEAIRM records the semantic distinctions required to understand such combinations, while:

  • AIPS remains authoritative for Product risk and Governance Obligation Profile computation;
  • M13 will model generic Evaluation/Assurance/Risk/Governance relationships;
  • GEAIRA may later define architectural safety and intervention responsibilities;
  • AIPDS remains responsible for operationalizing the applicable Product deployment controls.

M09-R29 - Governance non-duplication: GEAIRM MUST NOT create a second physical-AI risk tier, safety-obligation scheme or emergency-control taxonomy where AIPS already governs those Product semantics.


200. Canonical M09 Reference View​

                           +----------------------+
| Environment |
| GEAIRM-owned |
+----+-------------+---+
^ ^
| |
sensesEnvironment actsUponEnvironment
| |
+----------------+-------------+----------------+
| |
+----+-----+ +-----+------+
| Agent | | AI Execution|
| external | | GEAIRM-owned|
+----------+ +------------+

Sensing -------------------- distinct from ----------------> Perception
GEAIRM AIPS IC

Actuation ------------------ distinct from ----------------> PhysicalAction
GEAIRM AIPS IE

AIPS Product-Version Embodiment remains separate:

NonEmbodied
PhysicalWorldPerceptive
EmbodiedActuating
ClosedLoopAutonomous

Agent Ontology physical semantics remain separate:

RoboticAgent
SubstrateType: Living | Artificial | Hybrid

Control Loop
<---- participatesInControlLoop ---- Agent / AI Execution

Control Loop != ClosedLoopAutonomous
Control Loop != Agent Control Mode
Control Loop != Control Intelligence Capability

The view is semantic, not architectural. It does not prescribe sensor gateways, actuator services, robotics platforms, control buses, edge nodes or safety-controller topology.


201. Illustrative Semantic Examples​

201.1 Physical-world perception without actuation​

Environment Warehouse-Floor
classified as PhysicalEnvironment

AI Execution E-Vision
occursInEnvironment Warehouse-Floor
sensesEnvironment Warehouse-Floor
exercisesIntelligenceCapability Perception
producesOrCausesIntelligenceEffect DetectionAlert

AI Product Version P-Vision
hasEmbodiment PhysicalWorldPerceptive [AIPS]

No actuation or autonomous authority follows.

201.2 Robotic Agent using an actuating AI Product​

Robot-Agent R1
rdf:type RoboticAgent [Agent Ontology]
participatesInAIExecution E-Control

AI Product Version P-Control
hasEmbodiment EmbodiedActuating [AIPS]

AI Execution E-Control
occursInEnvironment Factory-Cell
actsUponEnvironment Factory-Cell
exercisesIntelligenceCapability Control
producesOrCausesIntelligenceEffect PhysicalAction

RoboticAgent and EmbodiedActuating remain independently asserted classifications.

201.3 Human-approved control loop​

Control Loop CL-1

AI Execution E-Recommend
participatesInControlLoop CL-1
exercisesIntelligenceCapability Perception
producesOrCausesIntelligenceEffect RecommendationEffect

Human Agent H1
participatesInControlLoop CL-1

AI Product Version P1
hasAgencyAutonomy ActWithApproval [AIPS]

The presence of a loop does not make the Product ClosedLoopAutonomous.

201.4 Digital feedback loop​

Environment Trading-Sandbox
classified as DigitalEnvironment

AI Execution E-Digital
occursInEnvironment Trading-Sandbox
participatesInControlLoop CL-Digital
producesOrCausesIntelligenceEffect DigitalAction

A control loop can be entirely digital and does not imply physical embodiment.

201.5 Sensor Tool without perceptive Product classification inference​

SensorService S1
plays Tool Role

Agent A1
usesTool S1

AI Execution E1
sensesEnvironment Environment-X

These statements alone do not infer AIPS PhysicalWorldPerceptive for any Product Version.


202. M09 Canonical Element Inventory​

ElementM03 kindAuthorityDispositionM09 status
EmbodimentClassificationAIPSREUSEDcanonical external
NonEmbodiedClassification valueAIPSREUSEDcanonical external
PhysicalWorldPerceptiveClassification valueAIPSREUSEDcanonical external
EmbodiedActuatingClassification valueAIPSREUSEDcanonical external
ClosedLoopAutonomousClassification valueAIPSREUSEDcanonical external
PerceptionIntelligence Capability valueAIPSREUSEDcanonical external
ControlIntelligence Capability valueAIPSREUSEDcanonical external
DigitalActionIntelligence Effect valueAIPSREUSEDcanonical external
PhysicalActionIntelligence Effect valueAIPSREUSEDcanonical external
RoboticAgentAgent typeKivaNura Agent OntologyREUSEDcanonical external
SubstrateTypeClassificationKivaNura Agent OntologyREUSEDcanonical external
EnvironmentConceptGEAIRMOWNEDcanonical
Environment KindClassificationGEAIRMOWNEDcanonical
PhysicalEnvironmentClassification valueGEAIRMOWNEDcanonical
DigitalEnvironmentClassification valueGEAIRMOWNEDcanonical
MixedEnvironmentClassification valueGEAIRMOWNEDcanonical
SensingConcept/activityGEAIRMOWNEDcanonical
ActuationConcept/activityGEAIRMOWNEDcanonical
Control LoopConcept / interaction patternGEAIRMOWNEDcanonical
occursInEnvironmentRelationship TypeGEAIRMOWNEDcanonical
sensesEnvironmentRelationship TypeGEAIRMOWNEDcanonical
actsUponEnvironmentRelationship TypeGEAIRMOWNEDcanonical
participatesInControlLoopRelationship TypeGEAIRMOWNEDcanonical

No executable GEAIRM identifiers are minted in M09. Stable machine identifiers remain deferred to the A1-A8 formalization programme after M17.


203. M09 Hard Distinctions​

The following distinctions are normative for subsequent slices:

AIPS Embodiment != Environment Kind
AIPS Embodiment != Agent Substrate Type
AIPS Embodiment != RoboticAgent identity

RoboticAgent != EmbodiedActuating automatically
Artificial Substrate != Artificial Intelligence
Artificial Substrate != AIPS Embodiment

Environment != AI Context
Environment != M03 Modelling Context
Environment != runtime platform

Sensing != Perception Intelligence Capability
Sensing != Artificial Intelligence
Actuation != Control Intelligence Capability
Actuation != PhysicalAction Intelligence Effect

DigitalAction != NonEmbodied automatically
PhysicalAction != EmbodiedActuating automatically
PhysicalAction != ClosedLoopAutonomous automatically

Control Loop != ClosedLoopAutonomous
Control Loop != Agent Control Mode
Control Loop != Control Intelligence Capability
Control Loop != autonomous authority

Tool use != Embodiment
physical capability != permission
actuation capability != authorization
closed-loop participation != autonomous authority

As in earlier slices, != denotes semantic non-equivalence rather than formal OWL disjointness unless later executable formalization explicitly authorizes stronger axioms.


204. M09 Normative Rules​

RuleRequirement
M09-R01GEAIRM MUST preserve AIPS authority for Embodiment and its controlled values.
M09-R02GEAIRM MUST preserve Agent Ontology authority for RoboticAgent and Agent substrate semantics.
M09-R03GEAIRM MUST NOT broaden AIPS hasEmbodiment beyond AI Product Version scope merely for generic environment modelling.
M09-R04Embodiment MUST remain distinct from AI Product Type, Product Family, AI Primitive Type and Agent subtype.
M09-R05Generic GEAIRM Control Loop MUST NOT be treated as equivalent to AIPS ClosedLoopAutonomous.
M09-R06PhysicalWorldPerceptive or EmbodiedActuating MUST NOT by itself determine autonomy, Control Mode or Delegated Authority.
M09-R07RoboticAgent identity and AIPS Embodiment MUST NOT be inferred from one another automatically.
M09-R08Agent Artificial Substrate Type MUST NOT be interpreted as Artificial Intelligence, AI Primitive identity or AIPS Embodiment.
M09-R09GEAIRM MUST NOT mint undocumented Agent-Ontology embodiment values merely from explanatory documentation.
M09-R10Environment MUST remain distinct from AI Context and M03 Modelling Context.
M09-R11Environment Kind MUST remain distinct from AIPS Embodiment.
M09-R12Interaction with a PhysicalEnvironment MUST NOT by itself establish RoboticAgent identity, EmbodiedActuating classification or PhysicalAction effect.
M09-R13occursInEnvironment MUST NOT imply Embodiment, Agent identity, physical co-location, autonomous authority or particular effects.
M09-R14Sensing MUST remain distinct from Perception and MUST NOT by itself establish Artificial Intelligence or AI Primitive status.
M09-R15Sensing MUST NOT automatically imply AI Context inclusion, Grounding, correctness or causal use.
M09-R16AIPS Control Intelligence Capability MUST remain distinct from Actuation.
M09-R17Actuation MUST remain distinct from PhysicalAction and MUST NOT be inferred solely from an effect classification.
M09-R18actsUponEnvironment MUST NOT by itself establish action-effect classification, autonomy, authorization, success or safety.
M09-R19DigitalAction MUST NOT by itself determine AIPS Embodiment or Agent identity.
M09-R20PhysicalAction MUST NOT by itself determine AIPS Embodiment, RoboticAgent identity, Control capability or autonomous authority.
M09-R21Control Loop MUST NOT by itself establish Artificial Intelligence, AI Primitive status or AI Product qualification.
M09-R22Control Loop MUST remain distinct from Agent Control Mode and AIPS Control Intelligence Capability.
M09-R23participatesInControlLoop MUST NOT imply control over every stage, autonomous authority, AgenticGoalDriven mode or ClosedLoopAutonomous Embodiment.
M09-R24Participation in a Control Loop MUST NOT be used to infer AIPS ClosedLoopAutonomous.
M09-R25Sensing, Perception, Control, Actuation, Intelligence Effect, Embodiment, Agent Control Mode and Delegated Authority MUST remain independently representable.
M09-R26GEAIRM 0.1 MUST NOT mint a universal Observation ontology where M07 context-resource semantics are sufficient.
M09-R27Use of sensing or actuation Tools MUST NOT by itself determine AIPS Embodiment.
M09-R28Physical capability, sensing, actuation, embodiment or loop participation MUST NOT by themselves establish permission or authorization to decide or act.
M09-R29GEAIRM MUST NOT duplicate AIPS physical-AI risk or safety-obligation taxonomies.
M09-R30Environment, sensing, actuation and Control Loop semantics MUST remain technology-neutral and MUST NOT prescribe robotics/runtime architecture.
M09-R31M09 relationships MUST follow M03 conservative inference defaults: no symmetry, transitivity, inverse or functional cardinality is implied unless explicitly stated.
M09-R32M09 semantic non-equivalence MUST NOT be strengthened into formal disjointness unless later executable formalization explicitly authorizes it.

205. M09 Exit Criteria​

GEAIRM-M09 is complete when:

  • AIPS Embodiment and all four current controlled values are explicitly reused without redefinition;
  • AIPS hasEmbodiment remains Product-Version scoped;
  • Embodiment is separated from Product Type, primitive type, Agent type and Environment Kind;
  • RoboticAgent and Agent Substrate Type remain governed by the Agent Ontology;
  • explanatory but non-machine-readable Agent embodiment values are not silently minted;
  • the M01 Environment semantic gap is resolved with a minimal generic Environment concept;
  • Physical, Digital and Mixed Environment kinds are distinguishable without replacing AIPS Embodiment;
  • AI Execution can be related to Environment without implying physical co-location or embodiment;
  • Sensing is defined and separated from AIPS Perception;
  • Actuation is defined and separated from AIPS Control and PhysicalAction;
  • DigitalAction and PhysicalAction remain Intelligence Effects rather than environment types or authority classifications;
  • generic Control Loop semantics are defined independently of AIPS ClosedLoopAutonomous;
  • control-loop participation can include human-supervised or non-AI arrangements;
  • physical capability remains distinct from permission/authorization;
  • physical-AI safety/risk governance remains under AIPS/M13 rather than duplicated;
  • no robotics/runtime architecture is prescribed;
  • M01-M08 authority, abstraction and conservative-inference rules remain intact.

M09 conclusion: these criteria are satisfied by this slice. GEAIRM now has a stable semantic model for physical and digital Environment interaction, sensing, actuation and feedback/control loops while preserving AIPS Embodiment, Intelligence Capability and Intelligence Effect as independent axes and preserving Agent Ontology authority for robotic/substrate semantics.


206. Forward Work​

The next slice is:

GEAIRM-M10 - Decision, Recommendation, Plan & Action Semantics

M10 will stabilize the enterprise-AI distinctions among inference/output, alert, recommendation, plan, decision formation, Decision, authorization and Action. It will reconcile AIPS Intelligence Effects with M08 Delegated Authority and M09 Environment interaction while preserving the boundary to GEARM/GEARA decision-support semantics and without treating a Decision effect or Action effect as automatic organizational authority.


207. GEAIRM-M10 - Decision, Recommendation, Plan & Action Semantics​

207.1 Purpose of this slice​

M09 established Environment interaction, Sensing, Actuation and Control Loop semantics while preserving Embodiment and physical-action classifications. M10 now establishes the semantic distinctions required to describe how enterprise AI may inform, recommend, plan, participate in decision formation, produce a Decision effect, receive or operate under Authorization, and cause Digital or Physical Action.

M10 answers eight questions:

  1. How should GEAIRM interpret informational Intelligence Effects such as Score, ClassificationEffect, PredictionEffect, DetectionAlert and ExplanationEvidence in relation to decision formation?
  2. How does RecommendationEffect differ from a Decision, Plan, Authorization or Action?
  3. How does Plan differ from a Decision and from execution of that plan?
  4. What generic concept is needed to represent Decision Formation without assuming that AI owns the decision?
  5. How should the AIPS Decision Intelligence Effect be interpreted without treating it as automatic legal, organizational or operational decision authority?
  6. What generic concept is needed to represent Authorization, and how does it differ from M08 Delegated Authority?
  7. How do DigitalAction and PhysicalAction relate to Decision, Authorization, M09 Environment interaction and Actuation without being collapsed into one mandatory pipeline?
  8. How does GEAIRM preserve the boundary to GEARM/GEARA and broader organizational decision/accountability semantics?

M10 does not define a universal legal theory of decision authority, an approval workflow architecture, an authorization service, a business-process engine or an organizational accountability model. Those concerns remain with applicable governance authorities, M13, GEAIRA and concrete enterprise implementations.


208. Source Semantics and Authority Resolution​

M10 is governed by the distinctions established in M01-M09.

AIPS already provides the following Intelligence Effect values relevant to this slice:

Score
ClassificationEffect
PredictionEffect
DetectionAlert
RecommendationEffect
GeneratedContent
ExplanationEvidence
Decision
Plan
DigitalAction
PhysicalAction

AIPS also provides the Product-Version Agency / Autonomy values:

Inform
Recommend
Assist
ActWithApproval
AutonomousAction

M06 established that Intelligence Effect answers what an AI Execution produces or causes, while M08 established that AIPS Agency / Autonomy answers what authority the AI Product Version exercises. M08 also introduced GEAIRM Delegated Authority as the bounded mandate within which an Agent may operate.

M09 further established that:

PhysicalAction != Actuation
PhysicalAction != autonomous authority
physical capability != permission

M10 preserves all of those semantics.

The M10 authority treatment is:

ConcernSemantic authorityM10 treatment
Score / ClassificationEffect / PredictionEffect / DetectionAlert / RecommendationEffect / GeneratedContent / ExplanationEvidenceAIPS Intelligence EffectREUSED
DecisionAIPS Intelligence EffectREUSED
PlanAIPS Intelligence EffectREUSED
DigitalAction / PhysicalActionAIPS Intelligence EffectREUSED
AIPS Agency / AutonomyAIPSREUSED
Delegated AuthorityGEAIRM M08REUSED
Decision FormationGEAIRMOWNED
AuthorizationGEAIRMOWNED
generic organizational/legal decision authorityapplicable enterprise/legal/governance authorityREFERENCED / outside M10 ownership
decision-support / analytical semanticsGEARM/GEARA and applicable authoritiesCROSS-DOMAIN-REFERENCE
approval/authorization architecture and policy enforcementGEAIRA / enterprise governance / implementationRESERVED

M10-R01 - AIPS effect authority preservation: GEAIRM MUST preserve AIPS authority for the existing Intelligence Effect values and MUST NOT redefine RecommendationEffect, Decision, Plan, DigitalAction or PhysicalAction as GEAIRM-owned replacements.

M10-R02 - AR authority preservation: GEAIRM MUST preserve AIPS authority for Agency / Autonomy and MUST NOT treat an Intelligence Effect as equivalent to an AR value merely because the terms appear related.


209. Informational and Evidential Effects​

M10 distinguishes informational/evidential effects from recommendation, decision and action semantics.

The following AIPS Intelligence Effects may provide material input to later reasoning or decision formation:

Score
ClassificationEffect
PredictionEffect
DetectionAlert
GeneratedContent
ExplanationEvidence

These effects may be significant, but their presence does not by itself establish:

  • a Recommendation;
  • a Plan;
  • a Decision;
  • Authorization;
  • delegated authority;
  • an Action;
  • organizational acceptance of the result;
  • correctness or truth.

For example:

PredictionEffect
may inform
Decision Formation

but

PredictionEffect != Decision

Likewise:

DetectionAlert != RecommendationEffect
DetectionAlert != Decision
ExplanationEvidence != Authorization
GeneratedContent != approved instruction

M10 therefore treats informational and evidential effects as possible contributors to decision formation, not as implicit decision or action authority.

M10-R03 - Informational-effect restraint: Score, ClassificationEffect, PredictionEffect, DetectionAlert, GeneratedContent and ExplanationEvidence MUST NOT by themselves be interpreted as Recommendation, Decision, Authorization or Action.


210. Recommendation Effect​

AIPS RecommendationEffect remains an Intelligence Effect value describing a recommendation produced by AI.

GEAIRM interprets Recommendation as advisory rather than determinative unless a separate authority relation establishes otherwise.

A Recommendation may:

  • propose an option;
  • rank candidate choices;
  • advise a human or Agent;
  • suggest a course of action;
  • contribute to Decision Formation.

A Recommendation does not by itself:

  • make the final Decision;
  • authorize the Decision;
  • authorize an Action;
  • prove that the recommendation was accepted;
  • establish AIPS Recommend Agency / Autonomy;
  • establish that a human reviewed it.

The distinction with AIPS AR is:

RecommendationEffect
What was produced?

Recommend
What authority does the AI Product Version exercise?

These may correlate in an AIPS Product profile, but M10 does not declare equivalence.

M10-R04 - Recommendation/Decision distinction: RecommendationEffect MUST NOT by itself establish a Decision Intelligence Effect or a final organizational decision.

M10-R05 - Recommendation/authority distinction: RecommendationEffect MUST NOT by itself establish Authorization, Delegated Authority or permission to act.

M10-R06 - Recommendation/AR distinction: AIPS RecommendationEffect and AIPS Agency / Autonomy value Recommend MUST remain semantically distinct.


211. Plan​

AIPS Plan remains an Intelligence Effect value.

M10 interprets a Plan as an effect representing a proposed or formed sequence, strategy, arrangement or course of steps intended to pursue an objective.

A Plan is distinct from both a Decision and an Action:

Plan
describes an intended course

Decision
represents a determination / selection

Action
changes or attempts to change a digital or physical state

A Plan may be:

  • generated before a Decision;
  • generated after a Decision;
  • revised during execution;
  • partially executed;
  • never executed;
  • used only as advisory material.

Therefore:

Plan != Decision
Plan != Authorization
Plan != Action
Plan != proof of execution

AIPS Planning Intelligence Capability and AIPS PLANNER AI Primitive Type remain separate under M04/M05:

Planning capability != Planner primitive != Plan effect

M10-R07 - Plan non-execution: Production of a Plan Intelligence Effect MUST NOT by itself imply that the Plan is accepted, authorized, executed or completed.

M10-R08 - Plan/Decision separation: Plan and Decision MUST remain distinct Intelligence Effects and MUST NOT be inferred from one another merely because a decision process uses a plan.


212. Decision Formation​

M10 introduces Decision Formation as the minimum generic concept needed to represent the process through which a determination may be formed without assuming that Artificial Intelligence is itself the final decision authority.

212.1 Canonical definition​

Decision Formation is the context-bounded process or activity through which relevant information, evidence, recommendations, plans, constraints, judgments and applicable authority are considered in forming a determination among available alternatives or courses.

Decision Formation may involve:

  • one or more Human Agents;
  • Artificial or Hybrid Agents;
  • one or more AI Executions;
  • analytical outputs;
  • Data or Information;
  • Recommendations;
  • Plans;
  • policies or Rules;
  • contextual constraints;
  • approval or authorization requirements.

Decision Formation is not intrinsically an AI process. AI may participate in it.

212.2 Element record​

FieldValue
Preferred labelDecision Formation
Element kindConcept / activity
DefinitionA context-bounded process or activity in which relevant information, evidence, recommendations, plans, constraints, judgments and applicable authority are considered in forming a determination.
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM10
Constraints / invariantsDecision Formation is not the AIPS Decision effect, Decision authority, Authorization, Agent identity or an architecture workflow.

Decision Formation does not imply that a Decision is ultimately produced. A process may terminate, escalate, defer or remain unresolved.

M10-R09 - Decision-formation/Decision distinction: Decision Formation MUST remain distinct from the AIPS Decision Intelligence Effect.

M10-R10 - Decision-formation/AI distinction: Participation of AI in Decision Formation MUST NOT by itself establish that AI is the final or accountable decision maker.


213. Participation in Decision Formation​

M10 introduces two generic relationships.

213.1 participatesInDecisionFormation​

FieldValue
Preferred labelparticipates in Decision Formation
DefinitionRelates an Agent or AI Execution to a Decision Formation process in which it materially participates in analysis, recommendation, reasoning, planning, judgment, approval, supervision or determination in the declared context.
Semantic authorityGEAIRM
Subject kindsAgent Ontology Agent or GEAIRM AI Execution
Object kindsGEAIRM Decision Formation
Directionalitydirected
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M10
Source sliceM10

Participation does not imply final decision authority or accountability.

213.2 informsDecisionFormation​

FieldValue
Preferred labelinforms Decision Formation
DefinitionRelates a semantic result, resource, Recommendation, Plan, evidence item or other relevant input to a Decision Formation process that may consider it.
Semantic authorityGEAIRM
Subject kindsapplicable effect occurrence, context resource, evidence/resource or other semantic input
Object kindsGEAIRM Decision Formation
Directionalitydirected
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M10
Source sliceM10

informsDecisionFormation means that the item is made relevant to the process. It does not imply acceptance, correctness, decisive causal influence or use in the final determination.

M10-R11 - Participation/authority restraint: participatesInDecisionFormation MUST NOT be interpreted as evidence of final decision authority, authorization or organizational accountability.

M10-R12 - Informing/acceptance restraint: informsDecisionFormation MUST NOT imply that the input is accepted, correct, decisive, authoritative or causally determinative.


214. Decision as an AIPS Intelligence Effect​

AIPS Decision remains an Intelligence Effect value answering what the AI Execution produces or causes.

M10 therefore interprets an AIPS Decision effect narrowly:

It records that a Decision-type effect was produced or caused by an AI Execution under the applicable AIPS semantics.

It does not by itself establish that the effect is:

  • legally binding;
  • organizationally final;
  • within Delegated Authority;
  • authorized for downstream action;
  • accepted by a human;
  • accountable to a particular person;
  • equivalent to a business-policy Decision;
  • automatically followed by an Action.

This distinction is essential:

Decision effect
!=
Decision authority

and:

Decision effect
!=
Authorization

A Decision effect may itself become an input to another Decision Formation process, such as a human review or higher-level approval.

M10-R13 - Decision-effect/authority separation: AIPS Decision MUST NOT by itself establish legal, organizational, operational or delegated decision authority.

M10-R14 - Decision-effect/action separation: Production of a Decision Intelligence Effect MUST NOT by itself imply DigitalAction, PhysicalAction or any other execution of that determination.


215. Authorization​

M08 introduced Delegated Authority as the bounded mandate under which an Agent may operate. M10 now introduces Authorization for a narrower semantic purpose.

215.1 Canonical definition​

Authorization is a context-bounded grant, approval or permission establishing that a specified Decision, AI Execution, invocation or Action may proceed within stated scope and conditions.

Authorization may be:

  • pre-existing;
  • granted at runtime;
  • conditional;
  • time-bounded;
  • action-specific;
  • resource-specific;
  • revocable;
  • dependent on human or system approval.

Authorization is not itself an architecture service, policy engine, entitlement store or workflow gate.

215.2 Element record​

FieldValue
Preferred labelAuthorization
Element kindConcept / permission assertion
DefinitionA context-bounded grant, approval or permission establishing that a specified Decision, AI Execution, invocation or Action may proceed within stated scope and conditions.
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM10
Constraints / invariantsAuthorization is not Delegated Authority, Agent Autonomy Level, Control Mode, Instruction, Recommendation, Decision effect, AR classification or implementation entitlement mechanism.

215.3 Authorization vs Delegated Authority​

The distinction is:

Delegated Authority
standing or bounded mandate within which an Agent may operate

Authorization
specific grant/approval/permission allowing a Decision,
invocation, Execution or Action to proceed within scope

An Authorization may be issued under or within a broader Delegated Authority, but GEAIRM does not require every Authorization to be modelled as derived from M08 Delegated Authority.

For example, a Human Agent with organizational authority may issue a runtime approval without the enterprise choosing to model that human's broader authority through the M08 Delegated Authority concept.

M10-R15 - Authorization/Delegated-Authority distinction: Authorization MUST remain distinct from Delegated Authority; neither MUST be treated as a synonym for the other.

M10-R16 - Authorization/autonomy distinction: Authorization MUST remain distinct from Agent Autonomy Level, Agent Control Mode and AIPS Agency / Autonomy.


216. Authorization Relationships​

216.1 authorizedBy​

M10 introduces a generic relationship for a concrete execution, Decision occurrence or Action occurrence that proceeds under a specific Authorization.

FieldValue
Preferred labelauthorized by
DefinitionRelates a context-bounded AI Execution, Decision occurrence, invocation or Action occurrence to an Authorization that permits it to proceed within stated scope and conditions.
Semantic authorityGEAIRM
Subject kindscontext-bounded AI Execution, Decision occurrence, invocation or Action occurrence
Object kindsGEAIRM Authorization
Directionalitydirected: occurrence -> Authorization
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M10
Source sliceM10

The relationship does not imply that:

  • the Authorization is sufficient for every related action;
  • the Authorization cannot expire or be revoked;
  • every condition attached to the Authorization has been satisfied;
  • the subject actually succeeded;
  • the subject is low risk;
  • the subject is autonomous.

216.2 withinDelegatedAuthority​

M10 introduces an optional relationship between a specific Authorization and a broader M08 Delegated Authority.

FieldValue
Preferred labelwithin Delegated Authority
DefinitionRelates an Authorization to a Delegated Authority whose bounded mandate provides or constrains the authority context within which the Authorization is valid.
Semantic authorityGEAIRM
Subject kindsGEAIRM Authorization
Object kindsGEAIRM Delegated Authority
Directionalitydirected
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M10
Source sliceM10

The relationship is optional because some authority sources may be governed externally rather than represented through M08 Delegated Authority.

M10-R17 - Authorization relationship restraint: authorizedBy MUST NOT imply success, safety, correctness, unrestricted scope, low risk or autonomous authority.

M10-R18 - Authority nesting restraint: withinDelegatedAuthority MUST NOT be interpreted as transitive transfer of every permission contained in the broader Delegated Authority.


217. AIPS Agency / Autonomy and Authorization​

AIPS AR and GEAIRM Authorization answer different questions.

AIPS Agency / Autonomy
What authority category does the AI Product Version exercise?

Authorization
Is this specific Decision / invocation / Execution / Action
permitted to proceed within this context and scope?

Examples:

217.1 Inform​

A Product Version classified Inform may produce informational effects. The classification does not itself authorize downstream action.

217.2 Recommend​

A Product Version classified Recommend may produce recommendations under AIPS semantics. A Recommendation still does not authorize downstream Decision or Action.

217.3 Assist​

Assist may permit more active support while still remaining distinct from specific authorization for material actions.

217.4 ActWithApproval​

ActWithApproval indicates the AIPS Product-Version authority category. In a concrete scenario, an approval may be represented through M10 Authorization, but GEAIRM does not redefine AIPS ActWithApproval as equivalent to Authorization.

217.5 AutonomousAction​

AutonomousAction permits autonomous-action behaviour under AIPS semantics. It does not mean globally unrestricted authority.

A Product Version classified AutonomousAction remains bounded by:

  • Product Promise and Product semantics;
  • applicable Delegated Authority;
  • policy and governance constraints;
  • Environment and resource scope;
  • risk obligations;
  • safety controls;
  • applicable Authorization conditions where required.

M10-R19 - AR/Authorization orthogonality: AIPS Agency / Autonomy classification MUST NOT substitute for context-specific Authorization.

M10-R20 - AutonomousAction scope restraint: AutonomousAction MUST NOT be interpreted as unlimited or globally unrestricted permission to decide or act.

M10-R21 - ActWithApproval mapping restraint: ActWithApproval MAY be associated with an Authorization/approval pattern in a concrete context but MUST NOT be declared semantically equivalent to GEAIRM Authorization.


218. Action Semantics​

M10 does not mint a universal standalone Action class merely to duplicate the AIPS Intelligence Effect dimension.

Instead, lowercase action is used as an editorial umbrella over concrete acts/interventions, while AI-caused action effects remain represented by the AIPS controlled values:

DigitalAction
PhysicalAction

This follows the same semantic minimality principle used in M04 for Intelligence.

218.1 DigitalAction​

DigitalAction indicates that the AI Execution produces or causes an action in a digital domain under AIPS Intelligence Effect semantics.

It does not by itself establish:

  • NonEmbodied;
  • Agent identity;
  • AgenticGoalDriven mode;
  • autonomous authority;
  • authorization;
  • successful completion;
  • a particular DigitalEnvironment.

218.2 PhysicalAction​

PhysicalAction indicates that the AI Execution produces or causes an action in the physical domain under AIPS Intelligence Effect semantics.

It remains distinct from:

  • M09 Actuation;
  • AIPS Embodiment;
  • RoboticAgent identity;
  • AIPS Control Intelligence Capability;
  • Authorization;
  • AIPS AutonomousAction.

218.3 Action occurrence vs effect classification​

AIPS DigitalAction/PhysicalAction are Intelligence Effect classifications. A concrete action occurrence may be classified using those effects where applicable.

M10 does not assert that the classification value itself is a persistent Action object or universal action ontology.

M10-R22 - No redundant Action ontology: GEAIRM 0.1 MUST NOT mint a universal Action superclass merely to duplicate AIPS DigitalAction/PhysicalAction Intelligence Effect semantics where the existing effect model is sufficient.

M10-R23 - Action/Authorization separation: DigitalAction or PhysicalAction MUST NOT by itself establish that the action was authorized.

M10-R24 - Action/success separation: An action effect or actsUponEnvironment assertion MUST NOT by itself establish successful completion, intended outcome or business acceptance.


219. Recommendation, Plan, Decision and Action Do Not Form a Mandatory Pipeline​

Enterprise AI scenarios are often drawn as:

Recommendation
->
Plan
->
Decision
->
Authorization
->
Action

M10 explicitly does not make this a mandatory universal sequence.

Valid enterprise scenarios may include:

Score -> human Decision

RecommendationEffect -> human Decision

Plan -> no Decision

Decision -> no Action

Decision -> later human approval -> Action

Control capability -> authorized PhysicalAction
without a separately represented Recommendation

DigitalAction
produced under pre-existing delegated authority
without a separate Plan effect

RecommendationEffect
used only for advisory information

Likewise, a Decision may be formed by a Human Agent without an AIPS Decision effect being produced by AI.

The semantic model is therefore a graph, not a fixed workflow.

M10-R25 - No mandatory decision pipeline: Recommendation, Plan, Decision, Authorization and Action MUST NOT be modelled as a universally mandatory sequential pipeline.

M10-R26 - No reverse inference: Presence of a later semantic element such as Action MUST NOT be used to infer that Recommendation, Plan or AI Decision necessarily occurred beforehand.


220. Decision Authority, Authorization and Accountability Are Separate​

M10 distinguishes three concepts that frequently collapse in implementations:

Decision effect
what AI produced

Decision authority
who/what is legitimately empowered to make the operative determination

Authorization
what specific Decision / invocation / Action is permitted to proceed

Accountability
who/what remains responsible under organizational, governance or legal arrangements

GEAIRM M10 owns only the minimal generic Authorization semantics and reuses M08 Delegated Authority. It does not define a complete generic DecisionAuthority or Accountability ontology.

Those meanings depend on organizational, regulatory, legal and governance context.

A Human Agent may remain accountable even where an AI Product is classified AutonomousAction. Conversely, a Human Agent approving an action does not automatically become Product owner or sole accountable party.

M10-R27 - Authority/accountability separation: Decision effect, Decision authority, Authorization and organizational/legal accountability MUST remain distinct semantic concerns.

M10-R28 - No accountability inference: GEAIRM MUST NOT infer organizational or legal accountability solely from Agent participation, Control Mode, Authorization, Decision effect or AIPS Agency / Autonomy.


221. Human Participation and Approval​

M08 established that:

HumanInLoop != ActWithApproval
HumanOnLoop != Recommend

M10 extends this principle.

A Human Agent may participate in Decision Formation as:

  • analyst;
  • reviewer;
  • recommender;
  • approver;
  • supervisor;
  • final decision maker;
  • action initiator.

Those are contextual participation roles and must not be inferred solely from Agent Control Mode.

Likewise, the presence of a Human Agent in Decision Formation does not prove that every AI result was reviewed or that every downstream action was approved.

Approval may be represented as an Authorization where the scenario requires a context-bounded grant to proceed.

M10-R29 - Human-participation restraint: Human participation in Decision Formation or a Control Loop MUST NOT by itself prove review, approval, final decision authority or accountability.


222. Decision and Action in Environment Interaction​

M09 established occursInEnvironment, sensesEnvironment, actsUponEnvironment, Actuation and Control Loop.

M10 aligns Decision/Action semantics with those concepts without collapsing them.

For example:

AI Execution
producesOrCausesIntelligenceEffect Decision

AI Execution
producesOrCausesIntelligenceEffect PhysicalAction

AI Execution
actsUponEnvironment Factory-Cell

These are three different assertions.

A Decision effect need not act upon an Environment. A PhysicalAction effect may be represented together with actsUponEnvironment, but the effect classification and environment-interaction relationship remain distinct.

Likewise:

PhysicalAction != Actuation
DigitalAction != DigitalEnvironment
Decision != Control Loop
Authorization != safety control

M10-R30 - Effect/environment orthogonality: Decision and Action Intelligence Effects MUST remain distinct from Environment-interaction relationships, Actuation and Control Loop participation.


223. Relationship to GEARM / GEARA and Decision Support​

GEAIRM M10 describes AI-specific semantic participation in recommendation, decision formation and action.

GEARM and GEARA remain separate authorities for Analytics / Information semantics and their architectural realization. Analytical or Information outputs may inform Decision Formation without becoming AI effects or AI Products merely because they contribute to the same enterprise decision.

Accordingly:

Analytical Output
may inform Decision Formation

Information Product
may inform Decision Formation

RecommendationEffect
may inform Decision Formation

none of those identities
implies the others

GEAIRM does not define the complete enterprise decision-support domain or replace GEARM/GEARA.

M10-R31 - Decision-support boundary: GEAIRM MUST interoperate with GEARM/GEARA decision-support semantics without absorbing the complete Analytics/Information decision domain.


224. Canonical M10 Reference View​

 Informational / Evidential Effects
Score | ClassificationEffect | PredictionEffect
DetectionAlert | GeneratedContent | ExplanationEvidence
\
\
v
+--------------------+
RecommendationEffect --->| Decision Formation |
Plan -------------------->| GEAIRM-owned |
Agent / AI Execution ---->| |
+----+----------+---------+
|
| may result in / contribute to
v
Decision effect
AIPS-owned
|
does not itself authorize
|
v
Authorization
GEAIRM-owned
|
authorizedBy / bounded by
|
+--------------------+--------------------+
| |
v v
DigitalAction PhysicalAction
AIPS IE AIPS IE
| |
| may be represented with | may be represented with
v v
Digital Environment / Physical or Mixed
digital interaction Environment
|
Actuation
M09

Agent
-- operatesUnderDelegatedAuthority --> Delegated Authority

Authorization
-- withinDelegatedAuthority -------> Delegated Authority

The diagram is conceptual. It does not prescribe a workflow engine, policy-decision point, approval service, orchestration runtime, human-task system or authorization architecture.


225. Illustrative Semantic Examples​

225.1 Fraud recommendation reviewed by a human​

AI Execution E1
producesOrCausesIntelligenceEffect Score
producesOrCausesIntelligenceEffect RecommendationEffect

RecommendationEffect occurrence R1
informsDecisionFormation DF1

Human Agent H1
participatesInDecisionFormation DF1

Nothing in these statements says that H1 accepted the Recommendation or that an Action occurred.

225.2 AI Decision effect without action authority​

AI Execution E2
producesOrCausesIntelligenceEffect Decision

AI Product Version P2
hasAgencyAutonomy Inform

The Decision effect does not override the Product Version's authority profile and does not authorize downstream action.

225.3 Plan approved for execution​

AI Execution E3
producesOrCausesIntelligenceEffect Plan

Plan occurrence P3
informsDecisionFormation DF3

Authorization A3
withinDelegatedAuthority DA3

Concrete downstream execution/action
authorizedBy A3

The Plan itself is not the Authorization.

225.4 Act-with-approval Product​

AI Product Version AP4
hasAgencyAutonomy ActWithApproval [AIPS]

Agent A4
operatesUnderDelegatedAuthority DA4 [GEAIRM]

Concrete action occurrence X4
authorizedBy AUTH4 [GEAIRM]

AUTH4
withinDelegatedAuthority DA4

ActWithApproval and AUTH4 are related at scenario level but are not the same semantic concept.

225.5 Autonomous physical action under bounded authority​

AI Product Version AP5
hasAgencyAutonomy AutonomousAction
hasEmbodiment ClosedLoopAutonomous

Agent A5
operatesUnderDelegatedAuthority DA5

AI Execution E5
participatesInControlLoop CL5
producesOrCausesIntelligenceEffect PhysicalAction
actsUponEnvironment Environment-5

AutonomousAction does not mean that DA5 is unlimited, nor that every physical action is automatically safe or authorized outside the bounded context.

225.6 Recommendation with no decision​

AI Execution E6
producesOrCausesIntelligenceEffect RecommendationEffect

The recommendation may be ignored, deferred or used only for awareness. No Decision is inferred.


226. M10 Canonical Element Inventory​

ElementM03 kindAuthorityDispositionM10 status
ScoreIntelligence Effect valueAIPSREUSEDcanonical external
ClassificationEffectIntelligence Effect valueAIPSREUSEDcanonical external
PredictionEffectIntelligence Effect valueAIPSREUSEDcanonical external
DetectionAlertIntelligence Effect valueAIPSREUSEDcanonical external
GeneratedContentIntelligence Effect valueAIPSREUSEDcanonical external
ExplanationEvidenceIntelligence Effect valueAIPSREUSEDcanonical external
RecommendationEffectIntelligence Effect valueAIPSREUSEDcanonical external
DecisionIntelligence Effect valueAIPSREUSEDcanonical external
PlanIntelligence Effect valueAIPSREUSEDcanonical external
DigitalActionIntelligence Effect valueAIPSREUSEDcanonical external
PhysicalActionIntelligence Effect valueAIPSREUSEDcanonical external
Agency / AutonomyClassificationAIPSREUSEDcanonical external
Delegated AuthorityConceptGEAIRM M08REUSEDcanonical
Decision FormationConcept / activityGEAIRMOWNEDcanonical
AuthorizationConcept / permission assertionGEAIRMOWNEDcanonical
Actioneditorial umbrellaGEAIRM editorial usagenot canonical standalone entityresolved
participatesInDecisionFormationRelationship TypeGEAIRMOWNEDcanonical
informsDecisionFormationRelationship TypeGEAIRMOWNEDcanonical
authorizedByRelationship TypeGEAIRMOWNEDcanonical
withinDelegatedAuthorityRelationship TypeGEAIRMOWNEDcanonical

No executable GEAIRM identifiers are minted in M10. Stable machine identifiers remain deferred to the A1-A8 formalization programme after M17.


227. M10 Hard Distinctions​

The following distinctions are normative for subsequent slices:

Score != RecommendationEffect
DetectionAlert != RecommendationEffect
ExplanationEvidence != Authorization

RecommendationEffect != Decision
RecommendationEffect != Authorization
RecommendationEffect != Action
RecommendationEffect != AIPS Recommend AR

Plan != Decision
Plan != Authorization
Plan != Action
Plan != proof of execution

Decision Formation != Decision Intelligence Effect
Decision effect != Decision authority
Decision effect != Authorization
Decision effect != Action
Decision effect != organizational accountability

Authorization != Delegated Authority
Authorization != Agent Autonomy Level
Authorization != Agent Control Mode
Authorization != AIPS Agency / Autonomy
Authorization != Instruction
Authorization != RecommendationEffect
Authorization != Decision effect

ActWithApproval != Authorization
AutonomousAction != unrestricted authority

DigitalAction != Authorization
PhysicalAction != Authorization
PhysicalAction != Actuation
Action effect != successful outcome

HumanInLoop != approval
Human participation != final decision authority automatically

Recommendation -> Decision -> Authorization -> Action
is not a mandatory universal pipeline

As in previous slices, != denotes semantic non-equivalence rather than formal OWL disjointness unless later executable formalization explicitly authorizes stronger axioms.


228. M10 Normative Rules​

RuleRequirement
M10-R01GEAIRM MUST preserve AIPS authority for the existing Intelligence Effect values relevant to recommendation, decision, plan and action.
M10-R02GEAIRM MUST preserve AIPS Agency / Autonomy as a distinct Product-Version authority classification.
M10-R03Informational/evidential effects MUST NOT by themselves be interpreted as Recommendation, Decision, Authorization or Action.
M10-R04RecommendationEffect MUST NOT by itself establish a Decision or final organizational determination.
M10-R05RecommendationEffect MUST NOT by itself establish Authorization, Delegated Authority or permission to act.
M10-R06RecommendationEffect and AIPS Recommend Agency / Autonomy MUST remain distinct.
M10-R07Plan MUST NOT by itself imply acceptance, Authorization, execution or completion.
M10-R08Plan and Decision MUST remain distinct Intelligence Effects.
M10-R09Decision Formation MUST remain distinct from the AIPS Decision Intelligence Effect.
M10-R10AI participation in Decision Formation MUST NOT by itself establish that AI is the final or accountable decision maker.
M10-R11participatesInDecisionFormation MUST NOT imply final decision authority, Authorization or accountability.
M10-R12informsDecisionFormation MUST NOT imply acceptance, correctness, decisiveness, authority or causal determination.
M10-R13AIPS Decision Intelligence Effect MUST NOT by itself establish legal, organizational, operational or delegated decision authority.
M10-R14Decision Intelligence Effect MUST NOT by itself imply DigitalAction, PhysicalAction or execution of the determination.
M10-R15Authorization MUST remain distinct from Delegated Authority.
M10-R16Authorization MUST remain distinct from Agent Autonomy Level, Agent Control Mode and AIPS Agency / Autonomy.
M10-R17authorizedBy MUST NOT imply success, safety, correctness, low risk, unrestricted scope or autonomous authority.
M10-R18withinDelegatedAuthority MUST NOT be interpreted as transitive transfer of every permission in the broader authority.
M10-R19AIPS Agency / Autonomy MUST NOT substitute for context-specific Authorization.
M10-R20AutonomousAction MUST NOT be interpreted as unlimited or globally unrestricted permission to decide or act.
M10-R21ActWithApproval MAY participate in an approval/Authorization pattern but MUST NOT be declared equivalent to GEAIRM Authorization.
M10-R22GEAIRM 0.1 MUST NOT mint a universal standalone Action ontology merely to duplicate AIPS DigitalAction/PhysicalAction semantics.
M10-R23DigitalAction or PhysicalAction MUST NOT by itself establish that the action was authorized.
M10-R24Action effect or Environment interaction MUST NOT by itself establish successful completion, intended outcome or business acceptance.
M10-R25Recommendation, Plan, Decision, Authorization and Action MUST NOT be modelled as a universally mandatory sequential pipeline.
M10-R26Presence of a later semantic element such as Action MUST NOT be used to infer that Recommendation, Plan or AI Decision necessarily occurred beforehand.
M10-R27Decision effect, Decision authority, Authorization and organizational/legal accountability MUST remain distinct.
M10-R28GEAIRM MUST NOT infer organizational/legal accountability solely from Agent participation, Control Mode, Authorization, Decision effect or AIPS AR.
M10-R29Human participation MUST NOT by itself prove review, approval, final decision authority or accountability.
M10-R30Decision and Action Intelligence Effects MUST remain distinct from Environment interaction, Actuation and Control Loop semantics.
M10-R31GEAIRM MUST interoperate with GEARM/GEARA decision-support semantics without absorbing the complete Analytics/Information decision domain.
M10-R32M10 relationships MUST follow M03 conservative inference defaults unless stronger semantics are explicitly established.
M10-R33M10 MUST NOT prescribe approval engines, policy-decision points, workflow systems, authorization services or other realization architecture.
M10-R34M10 semantic non-equivalence MUST NOT be strengthened into formal OWL disjointness unless later executable formalization explicitly authorizes it.

229. M10 Exit Criteria​

GEAIRM-M10 is complete when:

  • AIPS informational/evidential effects are separated from Recommendation, Decision, Authorization and Action;
  • RecommendationEffect is separated from AIPS Recommend Agency / Autonomy;
  • Plan is separated from Decision, Authorization and execution;
  • Decision Formation is defined as a generic process/activity without assuming AI final authority;
  • Agents and AI Executions can participate in Decision Formation without gaining implicit decision authority;
  • relevant resources/effects can inform Decision Formation without implying acceptance or correctness;
  • AIPS Decision remains an Intelligence Effect rather than a shortcut for legal/organizational Decision authority;
  • Authorization is minimally defined as a context-bounded grant/approval/permission;
  • Authorization is separated from Delegated Authority, autonomy and control-mode semantics;
  • specific occurrences can be related to Authorization without prescribing an authorization architecture;
  • Authorization can optionally be scoped within Delegated Authority;
  • AIPS ActWithApproval and AutonomousAction remain Product-Version authority classifications rather than generic Authorization objects;
  • DigitalAction and PhysicalAction remain AIPS Intelligence Effects;
  • no universal standalone Action ontology is introduced unnecessarily;
  • Decision/Action effects remain distinct from M09 Environment interaction and Actuation;
  • Recommendation, Plan, Decision, Authorization and Action are not forced into a universal pipeline;
  • human participation does not automatically imply approval, final authority or accountability;
  • GEARM/GEARA decision-support semantics remain adjacent rather than absorbed;
  • no approval/workflow/authorization architecture is prescribed;
  • M01-M09 authority, abstraction and conservative-inference rules remain intact.

M10 conclusion: these criteria are satisfied by this slice. GEAIRM now has a stable semantic model separating informational AI effects, recommendation, planning, decision formation, Decision effects, Authorization and digital/physical action while preserving AIPS Product authority classifications, M08 Delegated Authority, M09 Environment interaction and external organizational/accountability semantics.


230. Forward Work​

The next slice is:

GEAIRM-M11 - Dependency, Composition & Topology Model

M11 will define direct and transitive dependency, AI Primitive and Product dependency relationships, composition, hybrid-mechanism structure, multi-Agent participation, federation and topology while preserving the BPS principle that composition does not imply ontology inheritance and preserving AIPS distinctions among Primary/Derived AI Product classification, Composite derivation topology and broader Compound Product structure.


231. GEAIRM-M11 - Dependency, Composition & Topology Model​

231.1 Purpose of this slice​

M10 established Recommendation, Plan, Decision Formation, Authorization and Action semantics. M11 now establishes how enterprise-AI participants depend on, compose with, coordinate with and federate with other participants without transferring semantic identity through the graph.

M11 answers ten questions:

  1. How should GEAIRM reuse AIPS dependsOn and materiallyDependsOn without creating a second dependency vocabulary?
  2. What is the semantic difference between a direct dependency edge and a transitive dependency path?
  3. How should dependency paths be represented without declaring dependsOn itself universally transitive?
  4. How does material dependency differ from ordinary dependency, and is materiality transitive?
  5. What is a generic Composition, and why does composition not create Product, AI Product, Agent or AI Primitive identity by inheritance?
  6. How does generic composition differ from AIPS Composite derivation topology and BPS Compound Product?
  7. How do Hybrid Intelligence Engine, multi-Agent arrangements and runtime orchestration remain distinct from Product derivation?
  8. How should GEAIRM interpret AIPS Topology Scope without turning it into physical deployment topology?
  9. What generic meaning of Federation and Orchestration is needed for enterprise-AI relationships?
  10. How do version-specific dependency bindings and upstream change remain visible without pre-empting the detailed lifecycle/currentness model reserved for M14?

M11 is a semantic reference-model slice. It does not define service-mesh topology, network topology, deployment topology, orchestration engines, registries, workflow runtimes or dependency-resolution architecture. Those concerns belong to GEAIRA, AIPDS or implementation specifications.


232. Source Semantics and Authority Resolution​

M11 is governed by the semantic-authority rules of M01 and the composition non-inheritance principle already preserved through M03-M10.

The current AIPS core already establishes:

aips:dependsOn
General dependency relationship across Product, asset,
primitive, knowledge, tool and service resources.

aips:materiallyDependsOn
subPropertyOf aips:dependsOn
Dependency whose removal or non-equivalent replacement
would materially alter the essential intelligence proposition.

aips:derivedFrom
provenance-oriented derivation relationship.

AIPS qualification additionally requires a qualifying AI Product Version to have at least one direct or transitive AI Primitive path, and AIPS AIPrimitiveEvidence records a version-specific ordered dependencyPath.

AIPS also owns:

AI Product Type
Primary
Derived

Derivation Topology
SingleUpstream
Composite

Topology Scope
Standalone
DomainCoupled
CrossDomainFederated
EnterpriseOrchestrated

The AIPS taxonomy defines:

  • Primary — no material upstream AI Product dependency;
  • Derived — materially depends on one or more upstream AI Products;
  • Composite — a Derived derivation topology where multiple upstream AI Products materially contribute;
  • Topology Scope — a governance/risk dimension describing how far Product effects and coordination can propagate.

AIPDS further establishes that runtime composition/orchestration of several AI Products does not automatically create another AI Product. A new Derived AI Product exists only where the composition establishes a distinct governed intelligence proposition with its own Product identity/version and applicable AIPS semantics.

BPS remains authoritative for generic Product composition and Compound Product semantics.

M11 therefore adopts the following authority treatment:

ConcernSemantic authorityM11 treatment
dependsOnAIPSREUSED
materiallyDependsOnAIPSREUSED
derivedFromAIPS / PROV relationREUSED / REFERENCED
AIPS qualification dependencyPathAIPSREUSED at qualification-evidence scope
generic Dependency PathGEAIRMOWNED
CompositionGEAIRMOWNED as generic semantic arrangement
Compound ProductBPSREFERENCED
Primary / DerivedAIPSREUSED
SingleUpstream / CompositeAIPSREUSED
Topology ScopeAIPSREUSED
Hybrid Intelligence EngineAIPS AI Primitive TypeREUSED
Agent / Collective AgentKivaNura Agent OntologyREUSED
FederationGEAIRMOWNED as generic coordination pattern
OrchestrationGEAIRMOWNED as generic coordination pattern
runtime integration/deployment topologyAIPDS / GEAIRA / implementationRESERVED

M11-R01 - Dependency authority preservation: GEAIRM MUST reuse AIPS dependsOn and materiallyDependsOn where their existing semantics apply and MUST NOT mint semantically equivalent shadow dependency properties merely for local convenience.

M11-R02 - Product composition authority preservation: GEAIRM MUST preserve BPS Product/Compound Product authority and AIPS AI Product derivation semantics.


233. Dependency​

233.1 Dependency as a relationship rather than an intrinsic type​

A dependency is represented through a semantic relationship between participants. Dependency does not create a new intrinsic type for either endpoint.

GEAIRM reuses:

aips:dependsOn

for a general dependency assertion.

At generic reference-model level a dependency may involve, where semantically justified:

  • Products or Product Versions;
  • AI Primitives;
  • Models or Model Products;
  • Data or Data Products;
  • Information or Information Products;
  • Knowledge Resources;
  • Rules;
  • Tools;
  • Agents;
  • services or capabilities;
  • other governed resources.

The identity and authority of each endpoint remain governed independently.

233.2 Direct asserted dependency​

A single explicit dependsOn edge between a subject and object is treated by GEAIRM as a direct asserted dependency edge for that assertion.

This does not mean that AIPS dependsOn is formally constrained to direct-only use in every external dataset. It means GEAIRM will not infer unseen intermediate nodes from one asserted edge.

A -- dependsOn --> B

does not by itself say whether B depends on other resources.

233.3 Dependency does not imply ownership or containment​

A dependency may be remote, separately owned, externally governed, optional, replaceable or accessed through an interface.

Therefore:

A dependsOn B
!= A owns B
!= B is part of A's intrinsic identity
!= B is physically contained in A
!= A and B share Product Family

M11-R03 - Dependency non-inheritance: A dependency MUST NOT transfer Product Family, AI Product status, Agent identity, AI Primitive identity, risk tier, authority, governance ownership or other classifications from the dependency target to the dependent subject.

M11-R04 - Dependency/containment distinction: dependsOn MUST NOT by itself be interpreted as physical containment, ownership, co-deployment or component identity.


234. Material Dependency​

AIPS materiallyDependsOn is a subproperty of dependsOn and means that removal or non-equivalent replacement of the dependency would materially alter the essential intelligence proposition.

M11 reuses that meaning.

materiallyDependsOn
implies
dependsOn

but

dependsOn
does not imply
materiallyDependsOn

Materiality is assessed relative to the semantic proposition and context being evaluated.

For example:

  • a logging service may be operationally important but not material to the essential intelligence proposition;
  • an upstream AI Product may be material to a Derived AI Product's proposition;
  • a Tool may be replaceable by an equivalent Tool without materially changing the proposition;
  • a particular Knowledge Resource may be material in one Product/version and contextual in another.

234.1 Materiality is not universal criticality​

Material dependency does not automatically mean:

  • safety critical;
  • high risk;
  • non-replaceable;
  • sole dependency;
  • Product ownership;
  • architectural single point of failure.

Those are separate governance or architecture concerns.

234.2 Materiality is not automatically transitive​

Consider:

A materiallyDependsOn B
B materiallyDependsOn C

GEAIRM does not automatically infer:

A materiallyDependsOn C

The upstream role of C may or may not be material to A's essential intelligence proposition. Materiality must be established in the relevant context.

M11-R05 - Materiality non-transitivity: materiallyDependsOn MUST NOT be treated as universally transitive.

M11-R06 - Materiality/risk distinction: Material dependency MUST NOT by itself establish risk tier, criticality, irreversibility or architecture failure-domain semantics.


235. Dependency Path​

M11 introduces Dependency Path as the generic semantic construct needed to represent direct and transitive dependency reachability without declaring dependsOn itself transitive.

235.1 Canonical definition​

Dependency Path is an ordered sequence of one or more dependency edges connecting a source participant to a target participant through zero or more intermediate participants within a declared semantic and version context.

A path of one edge represents direct reachability:

A -> B

A path of more than one edge represents transitive reachability:

A -> B -> C -> D

235.2 Element record​

FieldValue
Preferred labelDependency Path
Element kindConcept / ordered relationship structure
DefinitionAn ordered sequence of one or more dependency edges connecting source and target through zero or more intermediate participants in a declared context.
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM11
Constraints / invariantsDependency Path is not Product composition, derivation topology, physical topology, ownership chain or automatic material-dependency chain.

235.3 Relationship to AIPS dependencyPath​

AIPS dependencyPath is scoped to AIPrimitiveEvidence and records the ordered version-specific path from a qualified AI Product Version to a qualifying AI Primitive.

GEAIRM does not broaden that AIPS property.

The relationship is:

GEAIRM Dependency Path
generic semantic construct

AIPS AIPrimitiveEvidence.dependencyPath
qualification-evidence representation of
a specific version-bound qualifying path

The AIPS property remains authoritative in its own evidence model.

235.4 Transitive dependency means path existence​

GEAIRM uses the phrase transitive dependency to mean that a dependency target is reachable through a Dependency Path containing more than one direct dependency edge.

It does not mint a new universal transitive transitivelyDependsOn property in M11.

This avoids unintended ontology reasoning where path semantics are context-, version- or materiality-sensitive.

M11-R07 - Path-based transitivity: GEAIRM MUST represent transitive dependency through explicit or derivable Dependency Path semantics rather than declaring AIPS dependsOn universally transitive.

M11-R08 - AIPS evidence-property preservation: GEAIRM MUST NOT broaden AIPS AIPrimitiveEvidence.dependencyPath beyond its qualification-evidence scope.


236. Version-Bound Dependencies​

AIPS qualification semantics require qualifying primitive dependency paths to be version-specific. AIPDS likewise strongly prefers versioned runtime dependency bindings for reproducibility and currentness analysis.

M11 therefore establishes a generic principle:

Where a dependency can change semantic qualification, behaviour, effect, authority, Product type or governance outcome, the dependency SHOULD be bound to a sufficiently precise version or governed identity for the applicable context.

M11 does not yet define the complete lifecycle/currentness model. That remains M14.

A statement such as:

ProductVersion-A
materiallyDependsOn ProductVersion-B

is semantically stronger and more reproducible than an unqualified dependency on a mutable label such as latest.

However, GEAIRM does not forbid non-versioned dependencies in every context. Some resources may be intentionally dynamic or externally resolved. The key requirement is that the semantic implications of mutability are not hidden.

M11-R09 - Version-sensitive dependency discipline: Dependency assertions that materially affect qualification or governed semantics SHOULD identify the applicable version or governed identity rather than rely on an ambiguous mutable target.

M11-R10 - No historical mutation inference: Publication of a new upstream version MUST NOT by itself be interpreted as rewriting the dependency graph of an existing downstream version that remains bound to an earlier dependency.


237. AI Product Type and Upstream Dependency​

M11 reuses AIPS AI Product Type rather than redefining Product derivation.

The controlling AIPS distinction remains:

Primary
no material upstream AI Product dependency

Derived
distinct intelligence proposition materially depends
on one or more upstream AI Products

This means:

AI Product Version
materiallyDependsOn upstream AI Product Version
may support
APT = Derived

subject to the authoritative AIPS qualification rules.

A dependency on a Model Product does not by itself make the Product Derived.

Likewise, dependency on:

  • Data Products;
  • Knowledge Resources;
  • Rules;
  • Tools;
  • AI Primitives that are not upstream AI Products;

does not by itself establish Derived APT.

M11-R11 - APT authority preservation: GEAIRM MUST NOT redefine Primary/Derived classification or infer Derived APT from dependencies other than the material upstream AI Product semantics established by AIPS.

M11-R12 - Model-Product boundary: Material dependency on a Model Product or AI Primitive MUST NOT by itself establish AIPS Derived AI Product classification.


238. Derivation Topology: SingleUpstream and Composite​

AIPS defines Derivation Topology for Derived AI Products:

SingleUpstream
Composite

Composite is not a third AI Product Type.

M11 preserves:

APT
Primary | Derived

Derivation Topology
SingleUpstream | Composite

A Derived AI Product with one material upstream AI Product may be classified SingleUpstream.

A Derived AI Product whose distinct intelligence proposition materially depends on multiple upstream AI Products may be classified Composite under AIPS semantics.

238.1 Composite is not generic composition​

The word Composite has a narrow AIPS meaning.

AIPS Composite
= Derived AI Product derivation topology

GEAIRM Composition
= generic semantic arrangement of participants

A generic Composition may exist without any Product at all.

M11-R13 - Composite/APT separation: AIPS Composite MUST remain a Derivation Topology value and MUST NOT be treated as a peer AI Product Type.

M11-R14 - Composite/generic-composition separation: Generic Composition MUST NOT be classified as AIPS Composite merely because multiple participants are present.


239. Composition​

M11 introduces Composition as a generic semantic arrangement independent of Product qualification.

239.1 Canonical definition​

Composition is a context-bounded arrangement in which two or more independently identifiable participants are intentionally combined, coordinated or used together to realize a broader behaviour, capability, interaction or proposition while retaining their independently governed identities unless another authority explicitly establishes a new composite identity.

Composition may involve:

  • AI Primitives;
  • Agents;
  • AI Products;
  • Model Products;
  • Tools;
  • Data/Information Products;
  • Knowledge Resources;
  • services;
  • Rules;
  • non-AI components.

Composition is not itself a Product.

239.2 Element record​

FieldValue
Preferred labelComposition
Element kindConcept / interaction arrangement
DefinitionA context-bounded arrangement in which two or more independently identifiable participants are intentionally combined, coordinated or used together to realize broader behaviour, capability, interaction or proposition.
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM11
Constraints / invariantsComposition is not Compound Product, AIPS Composite derivation, Hybrid Intelligence Engine, Collective Agent, AI Product qualification or architecture topology.

239.3 participatesInComposition​

FieldValue
Preferred labelparticipates in Composition
DefinitionRelates a participant to a Composition in which it is intentionally combined, coordinated or used with other participants in the declared context.
Semantic authorityGEAIRM
Subject kindsany applicable GEAIRM or externally governed participant
Object kindsGEAIRM Composition
Directionalitydirected: participant -> Composition
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M11
Source sliceM11

Participation in a Composition does not imply that the Composition owns the participant or that the participant's identity becomes intrinsic to every other participant.

M11-R15 - Composition non-inheritance: Participation in a Composition MUST NOT transfer Product Family, AI Product status, AI Primitive identity, Agent identity, authority, risk or other intrinsic classifications among participants.

M11-R16 - Composition non-productization: Composition MUST NOT by itself establish a new Product or AI Product identity.


240. Composition vs BPS Compound Product​

BPS owns generic Product ontology and the Compound Product concept.

A GEAIRM Composition is broader and may be purely operational, transient or non-Product.

A BPS Compound Product is a Product-level concept governed by Product ontology and its own identity/intrinsic-value semantics.

Therefore:

Composition
may participate in realizing a Product

but

Composition != Compound Product

A composition of Products does not automatically become a Compound Product. Product identity must be independently established under BPS.

Conversely, a Compound Product may be realized through many compositions over time without each runtime arrangement becoming a new Product identity.

M11-R17 - Composition/Compound-Product separation: GEAIRM Composition and BPS Compound Product MUST remain distinct; Product composition status MUST be established by BPS rather than inferred from a runtime or semantic arrangement.


241. Composition vs Derived AI Product​

AIPDS already establishes that orchestration of multiple AI Products is not automatically a new AI Product.

M11 makes the generic semantic rule explicit:

AI Product A
AI Product B
AI Product C
|
v
Composition / orchestration

does not automatically imply

Derived AI Product D

A new Derived AI Product requires the authoritative AIPS conditions, including a distinct governed intelligence proposition with its own Product identity/version and applicable qualification evidence.

Where such a new Product exists, the composition may be part of its constitution, but the Product identity is not created by graph shape alone.

M11-R18 - Runtime-composition/Product boundary: Runtime composition or orchestration of AI Products MUST NOT by itself establish a new Derived AI Product.

M11-R19 - Distinct-proposition requirement: Any claim that a Composition constitutes a new Derived AI Product MUST remain subject to AIPS qualification and Product-identity semantics, including the distinct governed intelligence proposition requirement.


242. Primitive Composition and Hybrid Intelligence Engine​

M05 established HYBRID_INTELLIGENCE_ENGINE as an AIPS AI Primitive Type and explicitly separated it from mere presence of multiple primitives.

M11 now allows multiple AI Primitives to participate in a Composition:

Primitive A ----\
Primitive B -----+--> Composition C
Primitive C ----/

This does not imply:

Composition C
classifiedAsPrimitiveType HYBRID_INTELLIGENCE_ENGINE

For the Hybrid Intelligence Engine classification to apply, the qualifying primitive itself must satisfy the AIPS mechanism semantics for that type.

A broader system composition may include learned, symbolic and knowledge-based primitives while still having no single primitive classified as Hybrid Intelligence Engine.

M11-R20 - Multi-primitive/hybrid distinction: Composition of multiple AI Primitives MUST NOT by itself establish HYBRID_INTELLIGENCE_ENGINE classification.

M11-R21 - Hybrid/Product distinction preservation: Hybrid Intelligence Engine classification MUST remain distinct from AIPS Composite derivation, BPS Compound Product and generic GEAIRM Composition.


243. Multi-Agent Composition​

M08 established that multiple Agents do not automatically form one CollectiveAgent.

M11 uses generic Composition to model multi-Agent arrangements:

Agent A ----\
Agent B -----+--> Composition M
Agent C ----/

The composition may represent cooperation, delegation, debate, supervision, handoff or other coordinated interaction.

No conclusion follows solely from multiple Agent participants that:

  • the group is a CollectiveAgent;
  • the composition is a Product;
  • the composition is an AI Product;
  • every Agent is Artificial;
  • every Agent has shared authority;
  • the composition has one unified memory, identity or lifecycle.

If the applicable Agent Ontology criteria establish Collective Agent identity, that is an independent Agent assertion.

M11-R22 - Multi-Agent/Collective-Agent distinction: Multiple Agents participating in a Composition MUST NOT automatically be classified as one Collective Agent.

M11-R23 - Multi-Agent/Product distinction: Multi-Agent Composition MUST NOT by itself establish Compound Product, AI Product or Composite Derived AI Product identity.


244. Federation​

M11 introduces Federation as a generic coordination pattern.

244.1 Canonical definition​

Federation is a context-bounded coordination arrangement in which independently governed participants cooperate or exchange capability, information, decisions or effects while retaining distinct identity and at least some independent authority, ownership or lifecycle.

Federation is intended to model semantic independence plus cooperation. It does not prescribe a network protocol, control plane or organization design.

Federation may involve:

  • Agents;
  • Products;
  • AI Products;
  • domain services;
  • Data/Information Products;
  • organizational or domain participants.

244.2 Element record​

FieldValue
Preferred labelFederation
Element kindConcept / coordination pattern
DefinitionA coordination arrangement among independently governed participants that cooperate while retaining distinct identity and at least some independent authority, ownership or lifecycle.
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM11
Constraints / invariantsFederation is not AIPS CrossDomainFederated Topology Scope, Collective Agent, Compound Product, physical network topology or automatic shared governance.

244.3 participatesInFederation​

FieldValue
Preferred labelparticipates in Federation
DefinitionRelates an independently governed participant to a Federation in which it cooperates while retaining its independent semantic identity.
Semantic authorityGEAIRM
Subject kindsapplicable Agent, Product, service, domain participant or other governed resource
Object kindsGEAIRM Federation
Directionalitydirected
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M11
Source sliceM11

Federation does not imply one central owner or absence of central coordination.

M11-R24 - Federation identity preservation: Participation in a Federation MUST NOT erase or merge participant identity, ownership, lifecycle or semantic authority.

M11-R25 - Federation/Collective distinction: Federation MUST NOT automatically establish Collective Agent or Compound Product identity.


245. Federation vs AIPS CrossDomainFederated​

AIPS CrossDomainFederated is a Topology Scope classification value used as a governance/risk driver for AI Product scope.

GEAIRM Federation is a generic semantic coordination arrangement.

Therefore:

Federation
generic coordination pattern

CrossDomainFederated
AIPS AI Product Topology Scope value

A Federation may exist without any participating Product being classified CrossDomainFederated.

Likewise, a Product may be classified CrossDomainFederated under AIPS based on its governed scope without GEAIRM asserting a first-class Federation resource.

M11-R26 - Federation/Topology-Scope separation: GEAIRM Federation and AIPS CrossDomainFederated MUST remain distinct and MUST NOT be inferred from one another solely from shared terminology.


246. Orchestration​

M11 introduces Orchestration as a generic coordination pattern.

246.1 Canonical definition​

Orchestration is a context-bounded coordination arrangement in which one or more participants direct, sequence, select or coordinate the invocation or interaction of other independently identifiable participants toward a broader objective or behaviour.

Orchestration may be:

  • static or dynamic;
  • AI-driven or non-AI;
  • Agent-led or workflow-led;
  • centralized or distributed;
  • Product-internal or cross-Product.

Orchestration does not itself establish Artificial Intelligence.

246.2 Element record​

FieldValue
Preferred labelOrchestration
Element kindConcept / coordination pattern
DefinitionA coordination arrangement in which one or more participants direct, sequence, select or coordinate other independently identifiable participants toward broader behaviour or an objective.
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM11
Constraints / invariantsOrchestration is not AI Product identity, Agent identity, AI Primitive identity, AIPS EnterpriseOrchestrated scope or architecture orchestration technology.

246.3 participatesInOrchestration​

FieldValue
Preferred labelparticipates in Orchestration
DefinitionRelates a participant to an Orchestration in which it coordinates or is coordinated in the declared context.
Semantic authorityGEAIRM
Subject kindsapplicable Agent, AI Execution, Product, AI Primitive, Tool, service or other participant
Object kindsGEAIRM Orchestration
Directionalitydirected
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M11
Source sliceM11

M11 does not introduce a universal orchestrator intrinsic type. An Agent may already be a StaticOrchestratorAgent under the Agent Ontology, while other orchestration may be performed by non-Agent workflows or services.

M11-R27 - Orchestration/AI distinction: Orchestration MUST NOT by itself establish Artificial Intelligence, AI Primitive status, Agent identity or AI Product qualification.

M11-R28 - Orchestration/Product distinction: Orchestration of multiple AI Products MUST NOT by itself establish a new AI Product.


247. Orchestration vs AIPS EnterpriseOrchestrated​

AIPS EnterpriseOrchestrated is a Topology Scope value used in Product governance/risk classification.

GEAIRM Orchestration is a generic coordination pattern.

Therefore:

Orchestration != EnterpriseOrchestrated

A local composition may use orchestration while the Product's Topology Scope remains Standalone or DomainCoupled under AIPS semantics.

Conversely, an AI Product may be classified EnterpriseOrchestrated because of its enterprise propagation/coordination scope even though M11 does not model a single Orchestration resource.

M11-R29 - Orchestration/Topology-Scope separation: GEAIRM Orchestration and AIPS EnterpriseOrchestrated MUST remain distinct and MUST NOT be inferred from one another automatically.


248. AIPS Topology Scope​

M11 reuses AIPS Topology Scope as an external Classification.

The current values are:

Standalone
DomainCoupled
CrossDomainFederated
EnterpriseOrchestrated

AIPS describes Topology Scope as a governance/risk driver concerning how far Product effects and coordination can propagate.

M11 therefore interprets it as governed Product scope, not as physical or deployment topology.

Topology Scope does not specify:

  • number of runtime instances;
  • network layout;
  • cloud region;
  • process boundaries;
  • service mesh;
  • deployment graph;
  • Agent count;
  • Product dependency cardinality;
  • Composition count.

248.1 Element record​

FieldValue
Preferred labelTopology Scope
Element kindClassification
DefinitionAIPS-controlled Product scope classification describing the governed propagation/coordination scope relevant to risk and governance.
Semantic authorityAIPS
Representation dispositionREUSED
Canonical valuesStandalone, DomainCoupled, CrossDomainFederated, EnterpriseOrchestrated
Constraints / invariantsTopology Scope is not physical topology, dependency graph shape, Federation identity, Orchestration identity or deployment architecture.

M11-R30 - Topology-Scope/architecture separation: AIPS Topology Scope MUST NOT be interpreted as physical deployment or infrastructure topology.

M11-R31 - Topology-Scope/graph-shape separation: Dependency count or graph shape MUST NOT by itself determine AIPS Topology Scope.


249. Semantic Topology Is a Graph View, Not a New Intrinsic Type​

M11 uses the term semantic topology editorially to refer to the arrangement of dependency, composition, federation, orchestration, Agent participation and Product relationships in a declared context.

M11 does not mint a standalone canonical SemanticTopology entity in GEAIRM 0.1.

The reasons are:

  1. AIPS already owns a specific TopologyScope classification.
  2. GEAIRM can represent topology-relevant semantics through explicit relationships and pattern concepts.
  3. A new generic topology class could be confused with architecture/deployment topology reserved for GEAIRA.
  4. Different views of the same semantic graph may legitimately emphasize different relationships without creating a new intrinsic identity.

Thus:

semantic topology
= editorial graph view over explicit semantics

not
a new Product/Agent/Primitive classification

M11-R32 - No redundant topology entity: GEAIRM 0.1 MUST NOT mint a universal Topology class merely to reify graph layout where dependency, Composition, Federation, Orchestration and AIPS Topology Scope already express the required semantics.


250. Dependency, Composition, Federation and Orchestration Are Orthogonal​

The four concerns answer different questions:

ConcernQuestion
DependencyWhat does this participant rely upon?
CompositionWhat participants are intentionally combined or used together?
FederationWhat independently governed participants cooperate while retaining identity?
OrchestrationWhat participants are coordinated or sequenced toward broader behaviour?
AIPS Topology ScopeHow far do governed Product effects/coordination propagate?

A single scenario may involve all five.

For example:

AI Product D
materiallyDependsOn AI Product A
materiallyDependsOn AI Product B

A and B
participateInComposition C

C
participates in Federation F

Agent O
participatesInOrchestration O1

AI Product D
hasTopologyScope CrossDomainFederated

No one assertion implies the others unless explicitly established.

M11-R33 - Orthogonality: Dependency, Composition, Federation, Orchestration and AIPS Topology Scope MUST remain independently representable.


251. Dependency Graphs Do Not Transfer Authority, Policy or Entitlement Automatically​

A dependency edge does not automatically propagate every semantic property of the target to the dependent participant.

In particular, M11 does not infer automatic propagation of:

  • authorization;
  • Delegated Authority;
  • Agent Autonomy Level;
  • Product ownership;
  • entitlement;
  • licensing rights;
  • risk tier;
  • Governance Obligation Profile;
  • source authority;
  • data classification;
  • safety approval;
  • Product Family.

AIPDS explicitly notes that policy inheritance across compositions is not assumed to be a simple union; conflicts require explicit governance resolution.

M11 preserves that principle at reference-model level.

M11-R34 - No blanket semantic propagation: Dependency or Composition MUST NOT be treated as an automatic propagation mechanism for authority, entitlement, policy, risk, governance obligations or Product classification.

M11-R35 - Policy-resolution boundary: Policy and governance propagation across dependencies/compositions MUST remain subject to explicit governance semantics rather than graph inheritance.


252. Dependency Change and Downstream Impact​

M11 recognizes that a change to an upstream dependency may affect downstream participants, but change impact is not the same as historical mutation.

A change may require reassessment of:

  • AI Product qualification;
  • APT;
  • Intelligence Capability or Effect;
  • Agency/Autonomy;
  • risk;
  • interface assumptions;
  • dependency currentness;
  • assurance.

However:

new upstream version
!= automatic downstream adoption

changed upstream semantics
!= automatic rewriting of prior downstream history

Only downstream versions whose actual bindings, dependency paths or governed assumptions are affected require corresponding impact analysis.

Detailed currentness, provenance and lifecycle semantics remain reserved for M14. M11 establishes only the dependency-graph principle.

M11-R36 - Impact/binding discipline: Upstream change MUST NOT be assumed to affect every downstream participant indiscriminately; impact depends on actual dependency bindings and material paths.

M11-R37 - Dependency-currentness deferral: M11 MUST NOT duplicate the detailed qualification/dependency currentness and lifecycle model reserved for AIPS and M14.


253. Canonical M11 Reference View​

                           DEPENDENCY

Participant A
|
+---- dependsOn ----------------------> Participant B
|
+---- materiallyDependsOn -----------> Participant C

Dependency Path:
A -> B -> C -> Primitive P
ordered path semantics
!= universal transitive property axiom


COMPOSITION

Participant A ----\
Participant B -----+---- participatesInComposition ---> Composition C
Participant C ----/

Composition C
!= Product automatically
!= Compound Product
!= AIPS Composite
!= Hybrid Intelligence Engine
!= Collective Agent


PRODUCT DERIVATION

AI Product Version D
materially depends on upstream AI Product(s)
|
v
AIPS APT = Derived
|
+-------+-------+
| |
SingleUpstream Composite

Composite
= AIPS Derivation Topology
!= generic Composition


FEDERATION

Participant A ----\
Participant B -----+---- participatesInFederation ----> Federation F
Participant C ----/

Federation F
!= CrossDomainFederated automatically
!= Collective Agent
!= Compound Product


ORCHESTRATION

Agent / workflow / execution / Product
|
+---- participatesInOrchestration ---> Orchestration O

Orchestration O
!= EnterpriseOrchestrated automatically
!= AI Product automatically


AIPS TOPOLOGY SCOPE

Standalone
DomainCoupled
CrossDomainFederated
EnterpriseOrchestrated

Product governance/risk scope
!= physical deployment topology

The view is semantic and deliberately independent of platform, network, runtime and service topology.


254. Illustrative Semantic Examples​

254.1 Primary AI Product with a Model Product dependency​

AI Product Version Fraud-A
dependsOn ModelProduct-M
materiallyDependsOn ModelProduct-M

ModelProduct-M
is an AI Primitive under AIPS

Fraud-A
APT = Primary

The Model Product dependency does not by itself make Fraud-A Derived.

254.2 Single-upstream Derived AI Product​

AI Product Version Intervention-D
materiallyDependsOn Fraud-A

Fraud-A
qualifies as upstream AI Product

Intervention-D
APT = Derived
DerivationTopology = SingleUpstream

The classification remains AIPS-owned.

254.3 Composite Derived AI Product​

AI Product Version Risk-D
materiallyDependsOn Fraud-A
materiallyDependsOn Sanctions-B

Risk-D
APT = Derived
DerivationTopology = Composite

Composite is not a peer APT.

254.4 Runtime composition with no new Product​

AI Product A
AI Product B
AI Product C

A, B, C
participateInComposition Runtime-C

No Derived AI Product is inferred because no distinct governed Product identity/proposition has been asserted.

254.5 Multi-Agent arrangement​

Agent A
Agent B
Agent C

A, B, C
participateInComposition Team-Run

No Collective Agent identity is inferred.

254.6 Federated cross-domain interaction​

Domain-A AI Product
Domain-B AI Product
Domain-C Information Product

all participateInFederation Federation-F

This does not automatically classify any AI Product as CrossDomainFederated; AIPS Topology Scope requires its own governed classification.

254.7 Orchestration is not Product creation​

Agent Orchestrator-A
participatesInOrchestration O1

AI Product P1
AI Product P2
Tool T1
participateInOrchestration O1

O1 is not automatically an AI Product, Compound Product or Derived AI Product.

254.8 Transitive primitive path​

Derived Product D
dependsOn Upstream Product U

U
dependsOn Model Product M

M
qualifies as AI Primitive

Dependency Path:
D -> U -> M

The path may support AIPS qualification evidence. GEAIRM does not declare dependsOn universally transitive.


255. M11 Canonical Element Inventory​

ElementM03 kindAuthorityDispositionM11 status
dependsOnRelationship TypeAIPSREUSEDcanonical external
materiallyDependsOnRelationship TypeAIPSREUSEDcanonical external
derivedFromRelationship TypeAIPS / PROVREUSED / REFERENCEDcanonical external
AIPS dependencyPathRelationship/evidence propertyAIPSREUSEDcanonical external at evidence scope
Dependency PathConcept / ordered relationship structureGEAIRMOWNEDcanonical
CompositionConcept / interaction arrangementGEAIRMOWNEDcanonical
participatesInCompositionRelationship TypeGEAIRMOWNEDcanonical
ProductConceptBPSREFERENCEDcanonical external
Compound ProductProduct conceptBPSREFERENCEDcanonical external
AI Product TypeClassificationAIPSREUSEDcanonical external
PrimaryClassification valueAIPSREUSEDcanonical external
DerivedClassification valueAIPSREUSEDcanonical external
Derivation TopologyClassificationAIPSREUSEDcanonical external
SingleUpstreamClassification valueAIPSREUSEDcanonical external
CompositeClassification valueAIPSREUSEDcanonical external
Hybrid Intelligence EngineAI Primitive Type valueAIPSREUSEDcanonical external
Collective AgentAgent typeKivaNura Agent OntologyREUSEDcanonical external
FederationConcept / coordination patternGEAIRMOWNEDcanonical
participatesInFederationRelationship TypeGEAIRMOWNEDcanonical
OrchestrationConcept / coordination patternGEAIRMOWNEDcanonical
participatesInOrchestrationRelationship TypeGEAIRMOWNEDcanonical
Topology ScopeClassificationAIPSREUSEDcanonical external
StandaloneClassification valueAIPSREUSEDcanonical external
DomainCoupledClassification valueAIPSREUSEDcanonical external
CrossDomainFederatedClassification valueAIPSREUSEDcanonical external
EnterpriseOrchestratedClassification valueAIPSREUSEDcanonical external
Semantic Topologyeditorial graph-view termGEAIRM editorial usagenot canonical standalone entityresolved

No executable GEAIRM identifiers are minted in M11. Stable machine identifiers remain deferred to the A1-A8 formalization programme after M17.


256. M11 Hard Distinctions​

The following distinctions are normative for subsequent slices:

dependsOn != materiallyDependsOn
direct dependency edge != transitive Dependency Path
Dependency Path != physical topology
material dependency != universal criticality
material dependency != automatically transitive

dependency != ownership
dependency != containment
dependency != Product inheritance
dependency != authority propagation

Composition != Product
Composition != Compound Product
Composition != AIPS Composite
Composition != Hybrid Intelligence Engine
Composition != Collective Agent

runtime composition != Derived AI Product automatically
Model Product dependency != Derived AI Product automatically

Primary != no dependencies
Derived != any upstream dependency
Composite != peer APT

multi-Agent Composition != Collective Agent
multi-Agent Composition != Compound Product
multi-Agent Composition != Composite Derived AI Product

Federation != CrossDomainFederated Topology Scope
Federation != Collective Agent
Federation != Compound Product

Orchestration != EnterpriseOrchestrated Topology Scope
Orchestration != Artificial Intelligence
Orchestration != Agent identity
Orchestration != AI Product identity

AIPS Topology Scope != deployment topology
AIPS Topology Scope != dependency graph shape

new upstream version != automatic downstream adoption
upstream change != historical mutation

As in previous slices, != denotes semantic non-equivalence rather than formal OWL disjointness unless later executable formalization explicitly authorizes stronger axioms.


257. M11 Normative Rules​

RuleRequirement
M11-R01GEAIRM MUST reuse AIPS dependsOn and materiallyDependsOn where their existing semantics apply and MUST NOT create shadow equivalents.
M11-R02GEAIRM MUST preserve BPS Product/Compound Product authority and AIPS Product derivation semantics.
M11-R03Dependency MUST NOT transfer Product Family, AI Product, Agent, AI Primitive, risk, authority or governance identity.
M11-R04dependsOn MUST NOT by itself imply containment, ownership, co-deployment or component identity.
M11-R05materiallyDependsOn MUST NOT be treated as universally transitive.
M11-R06Material dependency MUST NOT by itself establish risk tier, criticality or architecture failure-domain semantics.
M11-R07Transitive dependency MUST be represented through Dependency Path semantics rather than by declaring AIPS dependsOn universally transitive.
M11-R08GEAIRM MUST NOT broaden AIPS AIPrimitiveEvidence.dependencyPath beyond its qualification-evidence scope.
M11-R09Materially significant dependency assertions SHOULD identify the applicable version or governed identity where mutability can alter semantics.
M11-R10A new upstream version MUST NOT by itself rewrite the dependency graph of an existing downstream version.
M11-R11GEAIRM MUST preserve AIPS Primary/Derived authority and MUST NOT infer Derived from non-AI-Product dependencies.
M11-R12Model Product or AI Primitive dependency MUST NOT by itself establish Derived APT.
M11-R13AIPS Composite MUST remain a Derivation Topology value, not a peer AI Product Type.
M11-R14Generic Composition MUST NOT be classified as AIPS Composite merely because multiple participants are present.
M11-R15Participation in Composition MUST NOT transfer intrinsic identity, Product Family, Agent identity, AI Primitive identity, risk or authority.
M11-R16Composition MUST NOT by itself establish new Product or AI Product identity.
M11-R17GEAIRM Composition and BPS Compound Product MUST remain distinct.
M11-R18Runtime composition/orchestration of AI Products MUST NOT by itself establish a new Derived AI Product.
M11-R19Any claim that a Composition constitutes a Derived AI Product remains subject to AIPS qualification and distinct-governed-proposition semantics.
M11-R20Composition of multiple AI Primitives MUST NOT by itself establish Hybrid Intelligence Engine classification.
M11-R21Hybrid Intelligence Engine MUST remain distinct from AIPS Composite, BPS Compound Product and GEAIRM Composition.
M11-R22Multiple Agents participating in a Composition MUST NOT automatically be classified as one Collective Agent.
M11-R23Multi-Agent Composition MUST NOT by itself establish Compound Product, AI Product or Composite Derived AI Product identity.
M11-R24Federation participation MUST preserve participant identity, ownership, lifecycle and semantic authority.
M11-R25Federation MUST NOT automatically establish Collective Agent or Compound Product identity.
M11-R26GEAIRM Federation and AIPS CrossDomainFederated Topology Scope MUST remain distinct.
M11-R27Orchestration MUST NOT by itself establish Artificial Intelligence, AI Primitive status, Agent identity or AI Product qualification.
M11-R28Orchestration of multiple AI Products MUST NOT by itself establish a new AI Product.
M11-R29GEAIRM Orchestration and AIPS EnterpriseOrchestrated Topology Scope MUST remain distinct.
M11-R30AIPS Topology Scope MUST NOT be interpreted as physical deployment or infrastructure topology.
M11-R31Dependency count or graph shape MUST NOT by itself determine AIPS Topology Scope.
M11-R32GEAIRM 0.1 MUST NOT mint a universal Topology class merely to reify graph layout.
M11-R33Dependency, Composition, Federation, Orchestration and AIPS Topology Scope MUST remain independently representable.
M11-R34Dependency or Composition MUST NOT automatically propagate authority, entitlement, policy, risk, governance obligations or Product classification.
M11-R35Policy/governance propagation across dependency/composition MUST require explicit governance semantics rather than graph inheritance.
M11-R36Upstream change MUST NOT be assumed to affect every downstream participant; impact depends on actual bindings and material paths.
M11-R37M11 MUST NOT duplicate detailed dependency-currentness and lifecycle semantics reserved for AIPS and M14.
M11-R38M11 relationships MUST follow M03 conservative inference defaults unless stronger semantics are explicitly established.
M11-R39M11 MUST NOT prescribe dependency registries, orchestration engines, service meshes, deployment topology or federation infrastructure.
M11-R40M11 semantic non-equivalence MUST NOT be strengthened into formal OWL disjointness unless later executable formalization explicitly authorizes it.

258. M11 Exit Criteria​

GEAIRM-M11 is complete when:

  • AIPS dependsOn and materiallyDependsOn are explicitly reused rather than duplicated;
  • dependency is separated from ownership, containment and semantic inheritance;
  • material dependency is separated from ordinary dependency and universal criticality;
  • materiality is not assumed transitive;
  • direct and transitive dependency are represented through explicit edge/path semantics without making dependsOn universally transitive;
  • generic Dependency Path is defined while AIPS qualification dependencyPath remains evidence-scoped;
  • version-sensitive dependency semantics are explicit without pre-empting M14 lifecycle/currentness;
  • AIPS Primary/Derived Product Type is preserved;
  • Model Product / AI Primitive dependency does not automatically imply Derived;
  • AIPS SingleUpstream/Composite Derivation Topology is preserved;
  • generic Composition is defined independently of Product identity;
  • Composition is separated from Compound Product, Composite derivation, Hybrid Intelligence Engine and Collective Agent;
  • runtime composition/orchestration does not automatically create an AI Product;
  • multi-Agent arrangements can be represented without implying Collective Agent identity;
  • Federation is represented as independent-participant coordination without equating it to CrossDomainFederated scope;
  • Orchestration is represented generically without equating it to EnterpriseOrchestrated scope or Artificial Intelligence;
  • AIPS Topology Scope is reused as governed Product scope rather than deployment topology;
  • no redundant universal Topology entity is introduced;
  • dependency/composition graphs do not automatically propagate authority, policy, risk or entitlement;
  • upstream change impact is tied to actual bindings/material paths rather than historical mutation;
  • no dependency registry, service mesh, orchestration engine or physical/runtime topology is prescribed;
  • M01-M10 authority, abstraction and conservative-inference rules remain intact.

M11 conclusion: these criteria are satisfied by this slice. GEAIRM now has a stable semantic model for dependency edges and paths, material dependency, generic composition, Product derivation boundaries, multi-Agent arrangements, Federation, Orchestration and AIPS Topology Scope without allowing graph structure to transfer ontology identity or dictate architecture.


259. Forward Work​

The next slice is:

GEAIRM-M12 - Product Boundary & AIPS Bridge

M12 will connect the enterprise-AI reference model back to BPS Product ontology and AIPS AI Product semantics, positioning Product, AI Product, Model Product, AI-enabled Product, Compound Product, APT and qualification without allowing GEAIRM concepts such as AI Primitive, Agent, Composition, Tool, Decision Effect or Action Effect to become Products by inference.


260. GEAIRM-M12 - Product Boundary & AIPS Bridge​

260.1 Purpose of this slice​

M11 established dependency, composition, federation, orchestration and topology semantics without allowing graph structure to create Product identity. M12 now establishes the Product boundary between the generic enterprise-AI reference model and the BPS/AIPS Product ontology.

M12 answers ten questions:

  1. Which specification decides whether something is a Product at all?
  2. Which specification decides whether a Product belongs to the AI Product Family?
  3. How does GEAIRM Artificial Intelligence differ from an AIPS AI Product?
  4. How do AI Artifact/Asset, AI Primitive, AI Model, Model Product, AI Product and AI-enabled Product remain distinct?
  5. How does a BPS Compound Product interact with separately governed AI Product identity?
  6. How do Agent, Tool, Composition, RAG, Orchestration, Intelligence Effect and other GEAIRM concepts participate in Products without becoming Products by inference?
  7. How does AIPS AI Product Type bridge into GEAIRM dependency semantics without becoming a generic GEAIRM classification?
  8. How do AIPS Product-Version characteristics such as IC, MO, IE, AR, EM and Topology Scope relate to the generic GEAIRM semantic axes?
  9. How are Product qualification, AI Qualification Evidence and currentness kept under AIPS authority while GEAIRM remains usable for non-Product AI participants?
  10. How do AIPROD, AIPDS, GEAIRM and future GEAIRA remain complementary rather than overlapping specifications?

M12 deliberately introduces no new generic Product ontology. BPS remains authoritative for Productness and PIN. AIPS remains authoritative for AI Product Family qualification, AI Product Type, AI Qualification Evidence and AI Product-specific Product-Version semantics.


261. Semantic Authority and Jurisdiction for the Product Bridge​

The Product bridge follows the authority hierarchy already established in M01:

BPS
Productness
Product Intrinsic Nature (PIN)
Product Family
Product Kind
Contextual Product Role
Compound Product

AIPS
AI Product Family qualification
AI Qualification Evidence (AQE)
AI Product Type (APT)
AI Product Version intelligence/governance dimensions
AI-specific qualification currentness

GEAIRM
generic enterprise-AI concepts and relationships
whether or not those concepts are Products

GEAIRM therefore models the enterprise-AI domain across and beyond Products. It may represent AI Primitives, Agents, AI Executions, Context, Memory, Environments, Decision Formation, Tools, Composition and other semantic participants even when none of those things independently qualifies as a Product.

M12-R01 - Product authority preservation: GEAIRM MUST preserve BPS as the authority for Productness, PIN, Product Family, Product Kind, contextual Product roles and Compound Product semantics.

M12-R02 - AI Product authority preservation: GEAIRM MUST preserve AIPS as the authority for AI Product Family qualification, AQE, APT and AI Product-specific Product-Version classifications.


262. Productness Comes Before AI Product Qualification​

AIPS explicitly separates two questions:

1. Is this thing a Product?
-> BPS

2. If it is a Product, does its intrinsic nature
qualify it for the AI Product Family?
-> AIPS

This ordering is normative for the GEAIRM bridge.

A script, model checkpoint, prompt, dataset, transient workflow, Agent, AI Primitive, RAG composition or AI Execution does not become a Product merely because it is valuable, reusable, intelligent or governed in some other sense.

Conversely, a thing may be a valid Product while still not qualifying for the AI Product Family.

Examples include:

  • Model Product;
  • Model Service Product;
  • Feature Platform Product;
  • Knowledge Product;
  • Tool / Platform Product;
  • Service Product using AI;
  • Physical Product using AI;
  • Creative / Content Product constructed using AI.

M12-R03 - Qualification ordering: AI Product Family qualification MUST NOT be evaluated as a substitute for first establishing Productness under BPS.

M12-R04 - AI participation/Productness distinction: Participation in Artificial Intelligence, AI Execution, Agent activity, Tool use, Context, Composition, Federation or Orchestration MUST NOT by itself establish Productness.


263. Product Intrinsic Nature Remains the Product-Family Foundation​

BPS Product Intrinsic Nature (PIN) remains the framework used to establish what a Product fundamentally is.

The BPS dimensions remain:

PIN = Value-Bearing Essence
+ Essential Value Capability
+ Constitution
+ Identity Invariants

M12 does not reproduce or redefine those concepts.

The Product Family decision remains based on the Product's intrinsic value-bearing nature rather than merely on ingredients or dependencies.

Accordingly:

uses AI != AI Product Family
contains AI != AI Product Family
constructed using AI != AI Product Family
consumes AI Product != AI Product Family

A Movie created using generative AI remains a Creative / Content Product where the audiovisual work is the value-bearing essence. A streaming service using recommendation AI remains a Service Product where continuing service is the value-bearing essence. A smartphone containing AI remains a Physical / Engineered Product where the engineered device is the value-bearing essence.

M12-R05 - PIN primacy: GEAIRM MUST NOT infer Product Family from the presence, importance or number of AI dependencies where BPS PIN establishes another intrinsic Product Family.

M12-R06 - Composition non-inheritance reinforcement: Product constitution, dependency and composition MUST NOT automatically transfer AI Product Family membership to the containing or dependent Product.


264. Canonical AIPS AI Product Definition​

M12 reuses the AIPS AI Product definition:

An AI Product is a governed Product whose value-bearing essence is a distinct artificial-intelligence capability and whose current constitution directly or transitively contains at least one qualifying Artificial Intelligence Primitive.

This definition establishes a Product boundary around productized intelligence. It does not define Artificial Intelligence itself; M04 owns the generic GEAIRM Artificial Intelligence concept.

The distinction is:

Artificial Intelligence
generic enterprise-AI domain phenomenon
GEAIRM-owned

AI Product
governed Product whose intrinsic nature qualifies
for the AI Product Family
AIPS-owned

Therefore:

Artificial Intelligence != AI Product
AI Product != all Artificial Intelligence

An enterprise may contain extensive Artificial Intelligence without productizing every AI capability as an AI Product.

M12-R07 - AI/AI-Product distinction: GEAIRM Artificial Intelligence MUST remain distinct from AIPS AI Product identity.

M12-R08 - No universal productization: GEAIRM MUST NOT require every enterprise-AI capability, execution, mechanism or Agent to be productized as an AI Product.


265. AI Product Qualification Conditions AQ1-AQ5​

M12 reuses, but does not redefine, the five AIPS qualification conditions.

AQ1 - Product qualification​

The candidate first qualifies as a Product under BPS.

AQ2 - Intelligence primacy​

Artificial intelligence itself is the Product's value-bearing essence.

AQ3 - Distinct intelligence proposition​

The Product exposes an identifiable governed intelligence capability of its own.

AQ4 - Qualifying AI Primitive path​

The Product's current direct or transitive constitution contains at least one qualifying AI Primitive.

AQ5 - Governed Product boundary​

Ownership, lifecycle, contracts, controls, versioning, observability, evaluation and governance are defined for the Product boundary.

The conditions are conjunctive at AIPS level. AI Primitive presence is necessary but not sufficient.

A GEAIRM graph such as:

Agent
participatesInAIExecution E

E
involvesAIPrimitive P
exercisesIntelligenceCapability Reasoning

is therefore insufficient evidence that either the Agent, Execution or surrounding solution is an AI Product.

M12-R09 - AQ authority: GEAIRM MUST NOT weaken, replace or bypass AIPS AQ1-AQ5 by treating reference-model participation as equivalent AI Product qualification.

M12-R10 - Primitive necessary-not-sufficient: Presence of a qualifying AI Primitive path MUST NOT by itself establish AI Product status where Productness, intelligence primacy, distinct proposition or governed Product boundary are absent.


266. Distinct Intelligence Proposition​

AIPS requires a qualifying AI Product to expose a distinct governed intelligence proposition.

The Product-Version semantic record already carries AIPS intelligenceProposition; GEAIRM therefore does not mint a competing generic Intelligence Proposition class in M12.

At Product boundary, the proposition answers what distinct intelligence capability is intentionally being productized and governed as the Product's value proposition.

This is stronger than merely saying that a system invokes AI.

For example:

Router invokes Fraud AI Product
Router invokes Sanctions AI Product

without a new governed intelligence proposition
-> orchestration / composition
-> not automatically a new AI Product

while:

Transaction Risk Product
synthesizes upstream intelligence
exposes a distinct governed transaction-risk proposition
satisfies AQ1-AQ5
-> may qualify as a Derived AI Product

M12-R11 - Proposition/orchestration distinction: Routing, orchestration or aggregation of existing AI capabilities MUST NOT by itself establish the distinct intelligence proposition required by AIPS.

M12-R12 - Proposition authority preservation: GEAIRM MUST reuse the AIPS Product-Version intelligence-proposition semantics and MUST NOT create a competing Product-qualification proposition concept.


267. AI Qualification Evidence and Qualification Currentness​

AIPS AI Qualification Evidence (AQE) remains the evidence model proving why a specific Product version currently qualifies for the AI Product Family.

AQE includes, at minimum, evidence about:

  • qualification target;
  • value-bearing essence;
  • distinct intelligence proposition;
  • qualifying AI Primitive(s);
  • version-specific dependency path(s);
  • qualification status;
  • evaluation provenance.

The conceptual qualification statuses remain:

CURRENT
STALE
REVALIDATION_REQUIRED
INVALID

GEAIRM does not create a second qualification evidence model.

A Product identity may persist while one version is a current AI Product and a later version requires reclassification. Historical Product versions retain their historical evidence and classification.

M12-R13 - AQE non-duplication: GEAIRM MUST NOT create a second AI Product qualification-evidence structure where AIPS AQE already owns the semantics.

M12-R14 - Version-specific family membership: AI Product Family membership MUST be treated as version-specific and evidence-backed rather than a permanent label inherited by all versions of a Product identity.

M12-R15 - Invalid-currentness restraint: A Product Version whose authoritative AIPS qualification is INVALID MUST NOT be represented by GEAIRM as a currently qualified AI Product merely because an earlier version qualified.


268. AI Artifact / Asset vs Product​

AIPS distinguishes AI Artifacts/Assets from Products.

Examples of AI-related assets include:

  • raw model checkpoints;
  • datasets and evaluation sets;
  • embeddings;
  • prompt libraries;
  • feature definitions;
  • knowledge bases;
  • vector stores;
  • rulesets;
  • experimental pipelines;
  • benchmark results.

These resources may be critical to enterprise AI and may participate in GEAIRM Context, Memory, Retrieval, Grounding, Evaluation or dependency graphs without being Products.

An asset can later be productized, but Product identity requires BPS Product qualification.

M12-R16 - Asset/Product distinction: AI Artifact or Asset status MUST NOT be treated as Product status solely because the asset is governed, valuable, reusable or operationally important.


269. AI Primitive vs Product​

M04/M05 reuse AIPS AIPrimitive as the qualifying computational mechanism concept.

An AI Primitive may exist as:

  • a non-Product capability or component;
  • an AI Model;
  • a symbolic/knowledge/planning/search/optimization mechanism;
  • a Productized primitive such as a Model Product;
  • a dependency within an AI Product.

Therefore:

AI Primitive != Product
AI Primitive != AI Product

unless independent Product and AIPS qualification semantics establish those identities for the same real-world resource.

M12-R17 - Primitive/Product independent qualification: AI Primitive identity MUST NOT establish Product or AI Product identity without independent BPS/AIPS qualification.


270. Model Product vs AI Product​

AIPS defines a Model Product as a Product whose value-bearing essence is a reusable learned computational capability embodied in a governed model artifact or package.

A Model Product may simultaneously be:

  • a BPS Product;
  • an AIPS AI Primitive;
  • a vendor's contextual Business Product;
  • a procured or enabling Product for a consuming enterprise.

It is not automatically an AI Product.

The distinction is:

Model Product
productized learned computational capability
may serve as AI Primitive

AI Product
productized distinct governed intelligence proposition
satisfies AQ1-AQ5

A downstream AI assistant built from a procured LLM Model Product may still be Primary where it has no material upstream AI Product dependency.

M12-R18 - Model-Product/AI-Product separation: Model Product MUST remain distinct from AI Product; Productization of a learned model MUST NOT by itself establish AI Product Family membership.

M12-R19 - Model-Product/APT boundary: Dependency on one or more Model Products MUST NOT by itself make a qualified AI Product Derived.


271. AI-enabled Product​

An AI-enabled Product has another intrinsic Product Family while AI contributes materially to its construction or operation.

Representative examples include:

Smartphone + AI assistant
-> Physical / Engineered Product

Streaming Service + recommendation AI
-> Service Product

Movie created using generative AI
-> Creative / Content Product

AIPS explanatory semantics distinguish AI contribution roles such as:

  • constructional;
  • enabling;
  • essential.

M12 treats these as Product-constitution/boundary semantics governed by AIPS/BPS rather than as new GEAIRM Product classes.

A Product may therefore contain or consume substantial Artificial Intelligence while remaining outside the AI Product Family.

M12-R20 - AI-enabled non-inheritance: An AI-enabled Product MUST retain the Product Family established by its intrinsic nature unless BPS/AIPS qualification independently establishes a different or compound intrinsic classification.

M12-R21 - AI contribution role restraint: Constructional, enabling or essential AI contribution MUST NOT by itself be treated as AI Product Family qualification.


272. Compound Product and Co-essential AI​

BPS Compound Product semantics remain external to GEAIRM.

A cross-family Compound Product may contain AI as one of several co-essential intrinsic family contributions. The AIPS ontology uses a fully autonomous humanoid as a representative case where Physical / Engineered and AI contributions may both be essential.

This does not require the complete physical Product to collapse into a single AI Product identity.

A useful boundary is:

Humanoid Compound Product
Product identity / lifecycle / trust artifacts

Humanoid Intelligence AI Product
separately governed intelligence proposition
AIPS qualification / APT / AQE / AIPROD / AIPDS

The identities may be related but remain independently governed.

M12-R22 - Compound/AI-Product separation: A Compound Product with co-essential AI MUST NOT automatically be treated as identical to a separately governed AI Product participating in its constitution.

M12-R23 - Compound authority preservation: GEAIRM MUST NOT redefine the BPS criteria by which cross-family Compound Product identity is established.


273. Business Product Is a Contextual Role​

BPS treats Business Product as a contextual role rather than an intrinsic Product Family.

Therefore the same Product may be:

Model Product intrinsically
Business Product contextually for ModelVendor
Procured/Enabling Product contextually for EnterpriseConsumer

and an AI Product may similarly be:

AI Product intrinsically
Enabling Product internally for one enterprise
Business Product externally for another provider

Business Product role does not determine AI Product qualification, and AI Product Family membership does not determine Business Product role.

M12-R24 - Business-role/Product-family separation: Contextual Business Product, Enabling Product, Component Product, Procured Product or Utility/Platform roles MUST NOT be treated as intrinsic AI Product classifications.


274. Agent vs Product Boundary​

M08 preserves Agent identity under the KivaNura Agent Ontology.

An Agent may be:

  • a non-Product participant;
  • a component within a Product;
  • a Tool user;
  • an orchestrator;
  • independently productized under BPS;
  • a candidate AI Product if its Product boundary satisfies AIPS AQ1-AQ5.

An AI agent label therefore does not settle Product status.

Agent != Product
Agent != AI Product
ArtificialAgent != AI Product
AutonomousCognitiveAgent != AI Product

unless independent Product and AI qualification are established.

M12-R25 - Agent/Product separation: Agent identity or Agent subtype MUST NOT by itself establish Product or AI Product identity.


275. Tool Role vs Product Boundary​

M08 models Tool as a contextual Role.

A Tool-role resource may itself be:

  • an AI Product;
  • Model Product;
  • Service Product;
  • Data or Information resource;
  • AI Primitive;
  • API/service;
  • actuator or sensor;
  • non-Product capability.

Playing the Tool Role does not change Product identity.

Thus:

AI Product P
plays Tool Role in Agent execution E

P remains AI Product intrinsically
Tool remains a contextual Role

Likewise, a Tool does not become a Product merely because it is invoked repeatedly.

M12-R26 - Tool/Product independence: Tool Role MUST NOT create, erase or change Product Family, Product Kind or AI Product qualification of the underlying resource.


276. RAG, Composition and Orchestration vs Product Boundary​

M07 treats RAG as a compositional pattern rather than an intrinsic Product or primitive classification. M11 treats Composition and Orchestration as generic coordination arrangements.

Therefore:

RAG pattern != AI Product automatically
Composition != Product automatically
Orchestration != AI Product automatically

A RAG solution, Agent composition or orchestration layer may become a Product only where BPS Productness is independently established. It becomes an AI Product only where AIPS AQ1-AQ5 are also satisfied.

A distinct governed intelligence proposition is especially important: technical assembly of model + vector store + prompts + retrieval + tools does not by itself establish a Product boundary.

M12-R27 - Pattern/Product boundary: RAG, Composition, Federation or Orchestration MUST NOT be used as shortcuts for Product or AI Product qualification.


277. Intelligence Effect vs Product Boundary​

AIPS Intelligence Effects describe what intelligence produces or causes.

Examples include:

Score
ClassificationEffect
PredictionEffect
DetectionAlert
RecommendationEffect
GeneratedContent
ExplanationEvidence
Decision
Plan
DigitalAction
PhysicalAction

An effect is not automatically a Product.

A recurring recommendation, score, plan or generated output could potentially be incorporated into or independently productized through a governed Product proposition, but that requires a separate Product boundary and qualification decision.

Thus:

RecommendationEffect != Recommendation Product
Decision effect != Product
DigitalAction != Product
PhysicalAction != Product

unless BPS Productness is independently established for a distinct managed proposition.

M12-R28 - Effect/Product separation: Intelligence Effect occurrence or classification MUST NOT by itself establish Product, Information Product or AI Product identity.


278. AI Product Type Bridge​

AI Product Type remains an AIPS family-specific intrinsic classification and is only meaningful after AI Product Family qualification.

The values remain:

Primary
Derived

M11 already establishes the dependency interpretation:

Primary
no material upstream AI Product dependency

Derived
distinct intelligence proposition materially depends
on one or more upstream AI Products

M12 adds the Product-boundary constraint:

APT classifies the provenance/construction structure of productized intelligence; it does not classify generic GEAIRM Artificial Intelligence, AI Primitives, Agents, Executions, Compositions or Tools.

Likewise Composite remains a derivation topology for a qualified Derived AI Product, not a generic Product or GEAIRM Composition type.

M12-R29 - APT post-qualification rule: APT MUST NOT be assigned to a candidate that has not first qualified for the AI Product Family under AIPS.

M12-R30 - APT domain restraint: Primary/Derived MUST NOT be used as generic classifications for AI Primitives, Agents, AI Executions, Compositions or non-Product AI capabilities.


279. Product-Version AI Dimensions and GEAIRM Semantic Axes​

AIPS attaches governed characteristics to AIProductVersion. GEAIRM reuses several of the same authoritative vocabularies to describe generic enterprise-AI semantics at the appropriate reference-model level.

The bridge is:

AIPS Product-Version dimensionAIPS questionGEAIRM integration
APT - AI Product TypeWhere does productized intelligence originate?Product-only AIPS classification; dependency semantics integrated in M11/M12.
IC - Intelligence CapabilityWhat intelligence does the Product provide?Reused generically in M04/M06 for AI capability semantics.
MO - Execution ModeHow is Product intelligence invoked/operated?Reused in M06 for AI Execution occurrence semantics.
IE - Intelligence EffectWhat does intelligence produce/cause?Reused in M06/M10.
AR - Agency / AutonomyWhat authority does the Product Version exercise?Reused in M08/M10 while kept separate from Agent autonomy and Authorization.
EM - EmbodimentHow does Product intelligence relate to physical world?Reused in M09 while kept Product-Version scoped.
Topology ScopeHow far do Product effects/coordination propagate?Reused in M11 as governed Product scope, not deployment topology.
RT - Risk TierWhat AIPS governance regime applies?Remains AIPS-owned; integrated with generic risk/governance semantics in M13.

279.1 Product declaration vs execution occurrence​

AIPS Product-Version declarations and GEAIRM AI Execution assertions are complementary but not identical.

For example:

AI Product Version P
hasIntelligenceCapability Detection
hasExecutionMode RealTimeEventDriven
hasIntelligenceEffect DetectionAlert

states governed Product-Version semantics.

A particular GEAIRM execution occurrence may separately state:

AI Execution E
exercisesIntelligenceCapability Detection
operatesInExecutionMode RealTimeEventDriven
producesOrCausesIntelligenceEffect DetectionAlert

The Product declarations do not imply that every execution produces every declared effect, and one execution occurrence does not by itself reconstruct all Product metadata or qualification evidence.

M12-R31 - Product/occurrence separation: AIPS Product-Version characteristics and GEAIRM AI Execution occurrence assertions MUST remain semantically distinct and MUST NOT be treated as interchangeable records.

M12-R32 - No Product reconstruction from occurrence: GEAIRM execution assertions MUST NOT be used to infer complete AIPS Product qualification, Product Promise or Product-Version metadata.


280. AI Product Can Participate in GEAIRM Roles Without Losing Product Identity​

An AI Product or Product Version may participate in generic GEAIRM relationships while retaining its Product identity.

Examples include:

  • an AI Product is a dependency of another participant;
  • an AI Product plays the Tool Role for an Agent;
  • an AI Product participates in a Composition or Federation;
  • an AI Product provides effects consumed in Decision Formation;
  • an AI Product is part of a Control Loop;
  • an AI Product is an upstream dependency of a Derived AI Product.

These contextual relationships do not replace Product ontology.

Likewise, the Product may have its own AIPROD and AIPDS artifacts while participating in a wider GEAIRM semantic graph.

M12-R33 - Role preservation: GEAIRM contextual roles and relationships MUST NOT erase or replace the independently governed Product identity of participating Products.


281. GEAIRM Is Not an AI Product Qualification Engine​

GEAIRM provides the semantic reference model needed to understand enterprise AI participants and their relationships. It is not itself the authority that decides Product Family membership.

The responsibility split is:

BPS
Is it a Product?
What is its PIN / Product Family / contextual role?

AIPS AI Product Ontology + AQE
Does this Product Version qualify for the AI Product Family?
What is its APT and AI-specific Product classification?

GEAIRM
What generic enterprise-AI concepts participate,
and how do they relate regardless of Product status?

A future executable GEAIRM conformance runner must therefore validate GEAIRM model assertions, not independently certify AIPS Product qualification unless explicitly invoking an authoritative AIPS validation artifact.

M12-R34 - Qualification-engine boundary: GEAIRM conformance MUST NOT be represented as equivalent to AIPS AI Product qualification or AQE validation.


282. Relationship to AIPROD and AIPDS​

AIPS already separates semantic Product definition from deployment realization.

AI Product Ontology / Qualification
establishes what Product Version is and why it qualifies

AIPROD
records governed Product semantics:
qualification, APT, capabilities, dependencies,
governance, intent and lifecycle metadata

AIPDS
describes how that Product Version is operationalized,
served, secured, observed, scaled and evolved

GEAIRM sits alongside these artifacts as the generic enterprise-AI semantic reference model. It does not replace AIPROD or AIPDS.

For example, GEAIRM may define AI Context, Environment, Decision Formation or Composition; AIPROD may reference the Product's intended semantics; AIPDS may define the runtime realization. The three layers remain distinct.

M12-R35 - AIPROD boundary: GEAIRM MUST NOT replace the versioned Product semantic record owned by AIPROD.

M12-R36 - AIPDS boundary: GEAIRM MUST NOT prescribe or substitute for Product deployment and operational realization owned by AIPDS.


283. Relationship to Future GEAIRA​

M12 also preserves the hand-off to the future Generic Enterprise AI Reference Architecture.

The distinction remains:

GEAIRM
What are the generic enterprise-AI concepts and relationships?

AIPS Product semantics
When is a governed Product an AI Product?

GEAIRA
What architectural responsibilities are required to realize,
govern and interact with enterprise AI concepts and Products?

GEAIRA may later define architectural responsibilities for Product registries, execution, context/grounding, Agent/tool interaction, evaluation, serving, authority enforcement and observability. M12 does not pre-empt those architectural decisions.

M12-R37 - Architecture boundary: The Product bridge MUST remain semantic and MUST NOT prescribe GEAIRA planes, capabilities, ABBs, registries or implementation topology.


284. Canonical M12 Product Boundary View​

                              BPS
+-------------------+
| Candidate Thing |
+---------+---------+
|
qualifies as Product?
|
v
+-------------------+
| Product |
| PIN governed |
+---------+---------+
|
intrinsic Product Family
|
+-----------------+------------------+
| |
v v
another Product Family AIPS AI Qualification
Service / Physical / Data / AQ1 Productness
Creative / etc. AQ2 Intelligence primacy
| AQ3 Distinct proposition
| AQ4 AI Primitive path
| AQ5 Governed boundary
may be AI-enabled |
| v
| +-------------------+
| | AI Product |
| | AIPS-owned |
| +---------+---------+
| |
| version-specific AQE
| |
| APT after qualification
| Primary | Derived
| |
+------------------+------------------+
|
v
GEAIRM semantic graph

Artificial Intelligence | AI Primitive | Agent | AI Execution
AI Context | Memory | Tool Role | Environment | Control Loop
Decision Formation | Authorization | Dependency Path
Composition | Federation | Orchestration | Effects

GEAIRM concepts may participate in Products,
but participation does not create Product identity.

Special boundaries:

Model Product
Product + reusable learned computational capability
may be AI Primitive
!= AI Product automatically

AI-enabled Product
another intrinsic Product Family
AI contributes to construction/operation
!= AI Product automatically

Compound Product
BPS-governed cross-family Product
may include co-essential AI contribution
!= separately governed AI Product identity automatically

The view is conceptual and does not create new Product qualification semantics.


285. Illustrative Product-Boundary Examples​

285.1 Downloadable LLM sold by a model vendor​

Governed LLM
qualifies as BPS Product
intrinsic value = reusable learned computational capability
-> Model Product

Model Product
may be vendor Business Product
may be AI Primitive dependency
!= AI Product automatically

285.2 Enterprise assistant built from a procured Model Product​

Qwen Model Product
+ Enterprise Data / Knowledge / Rules / Tools
|
v
Corporate Assistant Product
distinct governed intelligence proposition
satisfies AQ1-AQ5
no material upstream AI Product dependency
|
v
AI Product
APT = Primary

The Model Product dependency does not force Derived.

285.3 Streaming service using recommendation AI​

Streaming Service Product
dependsOn Recommendation AI Product

PIN value-bearing essence
= continuing streaming service

Product Family
= Service Product

AI contribution
= enabling

The service does not inherit AI Product Family membership.

285.4 AI-generated movie​

Movie Product
constructed using Image / Video / Voice AI Products

PIN value-bearing essence
= audiovisual creative work

Product Family
= Creative / Content Product

Construction using AI does not make the Movie an AI Product.

285.5 Agent component vs Agent Product​

Agent A
participatesInAIExecution E
usesTool T

No Product boundary asserted
-> Agent participant, not Product by inference

If a separately managed Agent proposition is productized under BPS and satisfies AQ1-AQ5, that Product may independently qualify as an AI Product.

285.6 RAG solution​

Retrieval + Knowledge Resources + Prompt + AI Primitive
-> RAG composition

The graph alone does not answer Productness. A RAG solution may be a component or may become an AI Product only where a distinct governed Product boundary and intelligence proposition satisfy BPS/AIPS qualification.

285.7 Autonomous humanoid​

Humanoid Product
Physical / Engineered contribution
+ co-essential AI contribution
-> possible BPS Compound Product

Humanoid Intelligence capability
separately governed
-> may qualify as AI Product

The Compound Product and intelligence Product remain independently governable identities.

285.8 AI Product playing Tool Role​

Fraud Detection AI Product P
qualifies under AIPS

Within Agent execution E:
P plays Tool Role
Agent A usesTool P

Tool Role does not alter P's intrinsic Product identity.


286. M12 Canonical Element and Authority Inventory​

Element / concernM03 kindAuthorityDisposition in M12
ProductConceptBPSREFERENCED
Product Intrinsic NatureProduct ontology constructBPSREFERENCED
Value-Bearing EssencePIN dimensionBPSREFERENCED
Essential Value CapabilityPIN dimensionBPSREFERENCED
ConstitutionPIN dimensionBPSREFERENCED
Identity InvariantsPIN dimensionBPSREFERENCED
Product FamilyClassificationBPSREFERENCED
Product KindClassificationBPSREFERENCED
Contextual Product RoleRoleBPSREFERENCED
Business ProductContextual Product RoleBPSREFERENCED
Compound ProductProduct conceptBPSREFERENCED
AI ProductProduct Family conceptAIPS over BPSREUSED
AI Product VersionProduct Version conceptAIPSREUSED
AI Qualification EvidenceEvidence constructAIPSREUSED
intelligencePropositionProduct-Version semantic propertyAIPSREUSED
AI Product TypeClassificationAIPSREUSED
PrimaryClassification valueAIPSREUSED
DerivedClassification valueAIPSREUSED
SingleUpstreamDerivation Topology valueAIPSREUSED
CompositeDerivation Topology valueAIPSREUSED
Model ProductProduct / AI Primitive specializationAIPS over BPSREUSED
AI-enabled ProductProduct boundary conceptAIPS/BPS boundary semanticsREFERENCED
AI Artifact / AssetAsset conceptAIPSREFERENCED
AI PrimitiveConceptAIPSREUSED
Artificial IntelligenceConceptGEAIRM M04REUSED within GEAIRM
AgentConceptKivaNura Agent OntologyREUSED
ToolRoleGEAIRM M08REUSED
CompositionConcept / arrangementGEAIRM M11REUSED
FederationConcept / coordination patternGEAIRM M11REUSED
OrchestrationConcept / coordination patternGEAIRM M11REUSED
Intelligence EffectClassificationAIPSREUSED

M12 introduces no new executable identifiers. It is primarily a semantic-boundary and authority-bridge slice.


287. M12 Hard Distinctions​

The following distinctions are normative for subsequent slices:

Productness != AI Product qualification
Productized != AI Product
AI-related != Product
AI-related != AI Product

Artificial Intelligence != AI Product
AI Primitive != Product
AI Primitive != AI Product
AI Asset != Product automatically

Model Product != AI Product
Model Product dependency != Derived APT automatically

uses AI != AI Product
contains AI != AI Product
constructed using AI != AI Product
uses AI Product != AI Product

AI-enabled Product != AI Product automatically
Compound Product != AI Product automatically
Compound Product != separately governed AI Product identity

Business Product != Product Family
Business Product != AI Product qualification

Agent != Product
Agent != AI Product
Tool Role != Product Family
RAG != AI Product automatically
Composition != Product automatically
Orchestration != AI Product automatically

Intelligence Effect != Product
Decision effect != Product
DigitalAction != Product
PhysicalAction != Product

APT applies only after AI Product qualification
Primary / Derived != generic GEAIRM AI classification
Composite != generic Composition

AIPS Product-Version declaration != AI Execution occurrence
GEAIRM conformance != AIPS AI Product qualification
AIPROD != GEAIRM
AIPDS != GEAIRM
GEAIRA != GEAIRM

As in prior slices, != denotes semantic non-equivalence rather than formal OWL disjointness unless later executable formalization explicitly authorizes stronger axioms.


288. M12 Normative Rules​

RuleRequirement
M12-R01GEAIRM MUST preserve BPS authority for Productness, PIN, Product Family, Product Kind, contextual Product roles and Compound Product.
M12-R02GEAIRM MUST preserve AIPS authority for AI Product qualification, AQE, APT and AI Product-specific Product-Version semantics.
M12-R03AI Product Family qualification MUST NOT substitute for first establishing Productness under BPS.
M12-R04Participation in AI, Agent, Execution, Tool, Context, Composition, Federation or Orchestration semantics MUST NOT by itself establish Productness.
M12-R05Product Family MUST NOT be inferred from AI dependencies where BPS PIN establishes another intrinsic family.
M12-R06Product constitution/dependency/composition MUST NOT automatically transfer AI Product Family membership.
M12-R07GEAIRM Artificial Intelligence and AIPS AI Product MUST remain distinct.
M12-R08GEAIRM MUST NOT require every enterprise-AI capability, execution, mechanism or Agent to be productized.
M12-R09GEAIRM MUST NOT weaken, replace or bypass AIPS AQ1-AQ5 through reference-model participation.
M12-R10AI Primitive presence MUST NOT by itself establish AI Product status.
M12-R11Routing/orchestration/aggregation MUST NOT by itself establish the distinct intelligence proposition required by AIPS.
M12-R12GEAIRM MUST preserve AIPS Product-Version intelligence-proposition semantics rather than create a competing qualification concept.
M12-R13GEAIRM MUST NOT duplicate AIPS AI Qualification Evidence.
M12-R14AI Product Family membership MUST remain version-specific and evidence-backed.
M12-R15INVALID AIPS qualification MUST NOT be represented as current AI Product qualification merely because historical versions qualified.
M12-R16AI Artifact/Asset status MUST NOT be treated as Product status solely from value, reuse or governance.
M12-R17AI Primitive identity MUST NOT establish Product or AI Product identity without independent BPS/AIPS qualification.
M12-R18Model Product MUST remain distinct from AI Product and MUST NOT automatically inherit AI Product Family membership.
M12-R19Model Product dependency MUST NOT by itself establish Derived APT.
M12-R20AI-enabled Product MUST retain the Product Family established by intrinsic nature unless independent Product ontology semantics establish otherwise.
M12-R21Constructional, enabling or essential AI contribution MUST NOT by itself establish AI Product qualification.
M12-R22Compound Product with co-essential AI MUST NOT automatically be treated as identical to a separately governed AI Product.
M12-R23GEAIRM MUST preserve BPS authority for Compound Product identity.
M12-R24Contextual Business/Enabling/Component/Procured/Utility Product roles MUST NOT be treated as intrinsic AI Product classifications.
M12-R25Agent identity or subtype MUST NOT by itself establish Product or AI Product identity.
M12-R26Tool Role MUST NOT create, erase or change Product Family or AI Product qualification.
M12-R27RAG, Composition, Federation or Orchestration MUST NOT be used as shortcuts for Product or AI Product qualification.
M12-R28Intelligence Effect MUST NOT by itself establish Product, Information Product or AI Product identity.
M12-R29APT MUST NOT be assigned before AI Product Family qualification.
M12-R30Primary/Derived MUST NOT be used as generic classifications for AI Primitives, Agents, Executions, Compositions or non-Product AI capabilities.
M12-R31AIPS Product-Version classifications and GEAIRM AI Execution occurrence assertions MUST remain distinct.
M12-R32GEAIRM execution assertions MUST NOT reconstruct or substitute for complete AIPS Product qualification or Product-Version metadata.
M12-R33GEAIRM contextual roles and relationships MUST preserve independently governed Product identity.
M12-R34GEAIRM conformance MUST NOT be represented as equivalent to AIPS AI Product qualification or AQE validation.
M12-R35GEAIRM MUST NOT replace the versioned Product semantic record owned by AIPROD.
M12-R36GEAIRM MUST NOT prescribe or substitute for Product deployment realization owned by AIPDS.
M12-R37The Product bridge MUST remain semantic and MUST NOT prescribe GEAIRA planes, capabilities, ABBs, registries or implementation topology.
M12-R38M12 relationships and boundary statements MUST follow M03 conservative inference defaults unless stronger semantics are explicitly authorized.
M12-R39M12 semantic non-equivalence MUST NOT be strengthened into formal OWL disjointness unless later executable formalization explicitly authorizes it.

289. M12 Exit Criteria​

GEAIRM-M12 is complete when:

  • BPS remains the sole authority in this specification family for generic Productness and PIN;
  • AIPS remains authoritative for AI Product Family qualification, AQE, APT and Product-Version AI classifications;
  • Productness is explicitly evaluated before AI Product qualification;
  • GEAIRM Artificial Intelligence is separated from AI Product identity;
  • AQ1-AQ5 are preserved without being duplicated or weakened;
  • AI Qualification Evidence remains version-specific and AIPS-owned;
  • AI Asset/Artifact is separated from Product identity;
  • AI Primitive is separated from Product and AI Product identity;
  • Model Product is separated from AI Product while remaining able to act as Product + AI Primitive;
  • AI-enabled Product retains another intrinsic Product Family unless separately qualified;
  • Compound Product and separately governed AI Product identity remain distinct;
  • contextual Business Product role is separated from intrinsic Product Family;
  • Agent and Tool semantics remain independent of Product qualification;
  • RAG, Composition, Federation and Orchestration remain patterns/arrangements rather than qualification shortcuts;
  • Intelligence Effects remain effects rather than Products by inference;
  • APT is explicitly post-qualification and Product-only;
  • AIPS Product-Version characteristics are bridged to GEAIRM semantic axes without being confused with execution occurrences;
  • GEAIRM conformance remains distinct from AIPS Product qualification;
  • AIPROD and AIPDS responsibilities remain intact;
  • future GEAIRA architectural responsibility remains outside M12;
  • M01-M11 authority, abstraction and conservative-inference rules remain intact.

M12 conclusion: these criteria are satisfied by this slice. GEAIRM now has an explicit Product boundary and AIPS bridge: it can model enterprise AI broadly while preserving BPS Product ontology, AIPS AI Product qualification, Product-Version semantics, and the distinction between AI participation and governed productization.


290. Forward Work​

The next slice is:

GEAIRM-M13 - Evaluation, Assurance, Risk & Governance Model

M13 will model generic Evaluation, fitness, evidence, assurance, risk context, policy and control relationships while preserving AIPS authority for R0-R4 Risk Tier and Governance Obligation Profiles. It will distinguish evaluation results from assurance evidence, risk classification from governance obligations, and semantic governance concepts from future GEAIRA control/assurance architecture.


291. GEAIRM-M13 - Evaluation, Assurance, Risk & Governance Model​

291.1 Purpose of this slice​

M12 established the Product boundary and AIPS bridge. M13 now establishes the generic enterprise-AI semantics required to describe Evaluation, Evidence, Assurance, Risk and Governance without creating a second AIPS risk/governance framework or collapsing governance requirements into architecture/runtime controls.

M13 answers ten questions:

  1. What is a generic Evaluation in GEAIRM, and how does it differ from an AIPS AI Evaluation Report?
  2. What are Evaluation Criteria and Evaluation Results, and how can they be used without imposing one universal metric or PASS/FAIL vocabulary?
  3. How should the M01 AI Evidence semantic-gap candidate be resolved without introducing one artificial evidence superclass over all existing AIPS/PROV evidence types?
  4. What is the generic relationship between Evaluation and Assurance?
  5. How does generic Assurance differ from AIPS AI Product Assurance, COVP Fitness for Purpose and an AIPS Effective Risk Tier?
  6. How does GEAIRM reuse AIPS RiskDriverAssessment, Effective Risk Tier R0-R4 and Governance Obligation Profile without extending them to every non-Product AI participant?
  7. How are Governance Obligation, Policy, Rule, Instruction and Governance Control kept distinct?
  8. How is AIPDS Governance Control Binding positioned relative to semantic governance obligations?
  9. How can runtime evidence and continuous assurance trigger reassessment without allowing runtime observations to rewrite Product ontology directly?
  10. Which currentness, provenance and lifecycle semantics remain reserved for M14?

M13 deliberately does not define a universal enterprise risk taxonomy, legal/regulatory policy ontology, control library, assurance architecture, evaluation platform, model-monitoring stack, red-team platform or governance workflow. Those are external governance, GEAIRA, AIPDS or implementation concerns.


292. Source Semantics and Authority Resolution​

M13 is governed by M01 semantic authority and the boundaries established in M02-M12.

The current AIPS baseline already provides substantial evaluation, assurance, risk and governance semantics:

  • aips:AIEvalReport as an AIAsset;
  • aips:AIQualificationEvidence and its evaluationMethod;
  • Product-level AI Product Assurance in the AIPS DPP trust profile;
  • COVP contextual Fitness for Purpose;
  • aips:EffectiveRiskTier;
  • aips:RiskTierLevel with R0-R4;
  • aips:RiskDriverAssessment and aips:RiskDriverKind;
  • aips:GovernanceObligation;
  • aips:GovernanceObligationProfile;
  • AIPDS aips:GovernanceControlBinding;
  • Product/deployment governance and continuous-assurance rules.

M13 therefore uses the following authority treatment:

ConcernSemantic authorityM13 treatment
AI Evaluation ReportAIPSREUSED
AI Qualification Evidence / evaluationMethodAIPSREUSED at qualification scope
generic Evaluation activityGEAIRMOWNED
Evaluation CriterionGEAIRMOWNED
Evaluation ResultGEAIRMOWNED
EvidenceGEAIRM contextual Role over externally governed resourcesOWNED as Role
AI Product AssuranceAIPS DPP / AIPS trust semanticsREUSED / REFERENCED
generic Assurance AssessmentGEAIRMOWNED
Assurance FindingGEAIRMOWNED
Fitness for Purpose / Fitness StatusAIPS COVPREUSED / REFERENCED
Effective Risk Tier / R0-R4AIPSREUSED
Risk Driver Assessment / Risk Driver KindAIPSREUSED
Governance Obligation / Governance Obligation ProfileAIPSREUSED
Policyapplicable enterprise/regulatory governance authorityREFERENCED; no universal GEAIRM policy ontology
Governance Controlapplicable enterprise control framework / AIPDS realizationREFERENCED / RESERVED
Governance Control BindingAIPDSREUSED at deployment-realization scope
assurance/risk/control architectureGEAIRA / implementationRESERVED

M13-R01 - AIPS governance authority preservation: GEAIRM MUST preserve AIPS authority for Effective Risk Tier, R0-R4, Risk Driver Assessment, Governance Obligation and Governance Obligation Profile.

M13-R02 - Product assurance authority preservation: GEAIRM MUST preserve AIPS authority for AI Product Assurance and COVP Fitness for Purpose where Product-version trust and consumption semantics are concerned.


293. Evaluation​

M01 identified generic Evaluation as a semantic-gap candidate because AIPS provides Evaluation Reports, evaluation methods, metrics and evaluation-related governance obligations without establishing one generic enterprise-AI Evaluation activity.

M13 resolves the gap by introducing a minimal GEAIRM Evaluation concept.

293.1 Canonical definition​

Evaluation is a context-bounded assessment activity in which a declared target is examined against one or more explicit criteria using stated methods and available evidence in order to produce one or more Evaluation Results.

Evaluation is deliberately mechanism-neutral. It may evaluate:

  • an AI Primitive;
  • an Intelligence Capability;
  • an AI Execution;
  • an Agent behaviour;
  • a Product or AI Product Version;
  • a Composition;
  • an Environment interaction;
  • an Authorization pattern;
  • a safety property;
  • a governance property;
  • another explicitly declared target.

Evaluation may be quantitative, qualitative, deterministic, statistical, simulation-based, scenario-based, human-reviewed or hybrid.

293.2 Element record​

FieldValue
Preferred labelEvaluation
Element kindConcept / assessment activity
DefinitionA context-bounded assessment activity examining a declared target against explicit criteria using stated methods and available evidence to produce Evaluation Results.
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM13
Constraints / invariantsEvaluation is not an AI Evaluation Report, Assurance, Risk Tier, Governance Obligation, governance Control, AI Qualification or architecture capability.

293.3 Evaluation is not necessarily an AI capability​

The act of evaluating an AI participant may itself be carried out by:

  • a human assessor;
  • deterministic validation;
  • statistical analysis;
  • an AI-assisted evaluation mechanism;
  • a hybrid process.

Therefore:

Evaluation
!= Artificial Intelligence automatically
!= AI Primitive automatically
!= Intelligence Capability automatically

If an evaluation mechanism independently satisfies AI semantics, that identity must be asserted separately.

M13-R03 - Evaluation/AI distinction: Conducting Evaluation MUST NOT by itself establish Artificial Intelligence, AI Primitive or AI Product identity for the evaluator.


294. Evaluation Target​

M13 does not mint a universal intrinsic EvaluationTarget class. Instead, Evaluation Target is a contextual Role played by the thing being assessed in a particular Evaluation.

This follows the role discipline established in M03, M07 and M08.

A resource may be:

  • an AI Primitive intrinsically;
  • a Product under BPS;
  • an Agent under the Agent Ontology;
  • an Environment under GEAIRM;
  • and an Evaluation Target in one Evaluation.

Playing the Evaluation Target Role does not change the participant's identity.

294.1 evaluatesTarget​

FieldValue
Preferred labelevaluates Target
DefinitionRelates an Evaluation to a participant or property being assessed in the declared evaluation context.
Semantic authorityGEAIRM
Subject kindsGEAIRM Evaluation
Object kindsany applicable participant/property playing Evaluation Target Role
Directionalitydirected
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M13
Source sliceM13

An Evaluation may have multiple targets where explicitly declared.

M13-R04 - Evaluation-target role restraint: Playing the Evaluation Target Role MUST NOT alter Product Family, Agent identity, AI Primitive identity, risk classification or governance ownership.


295. Evaluation Criterion​

295.1 Canonical definition​

Evaluation Criterion is a declared condition, expectation, threshold, question, property or benchmark against which an Evaluation Target is assessed within a stated context.

Evaluation Criterion is generic. It may refer to:

  • accuracy;
  • robustness;
  • calibration;
  • logical consistency;
  • groundedness;
  • safety;
  • latency;
  • reversibility;
  • policy compliance;
  • fairness;
  • explanation quality;
  • environment coverage;
  • intended-use fitness;
  • another declared property.

M13 does not standardize the criteria themselves. The applicable Product, domain, policy, regulatory or evaluation authority owns their substantive meaning.

295.2 Element record​

FieldValue
Preferred labelEvaluation Criterion
Element kindConcept
DefinitionA declared condition, expectation, threshold, question, property or benchmark against which a target is assessed.
Semantic authorityGEAIRM for generic role/structure; criterion substance may be externally governed
Representation dispositionOWNED
Normative statuscanonical
Source sliceM13
Constraints / invariantsCriterion is not Evaluation Result, Policy, Rule, Governance Obligation or control by default.

295.3 usesEvaluationCriterion​

FieldValue
Preferred labeluses Evaluation Criterion
DefinitionRelates an Evaluation to an Evaluation Criterion applied in that assessment.
Semantic authorityGEAIRM
Subject kindsEvaluation
Object kindsEvaluation Criterion
Directionalitydirected
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M13
Source sliceM13

M13-R05 - Criterion/governance distinction: Evaluation Criterion MUST NOT be treated as Policy, Rule, Governance Obligation or Governance Control merely because the criterion reflects one of those sources.


296. Evaluation Result​

296.1 Canonical definition​

Evaluation Result is a context-bounded finding, measurement, observation, score, status or conclusion produced by an Evaluation against one or more declared criteria.

M13 intentionally does not impose one universal result vocabulary.

An Evaluation Result may be:

  • numerical;
  • categorical;
  • narrative;
  • multi-metric;
  • scenario-specific;
  • criterion-specific;
  • uncertain or inconclusive.

296.2 Element record​

FieldValue
Preferred labelEvaluation Result
Element kindConcept
DefinitionA context-bounded finding, measurement, observation, score, status or conclusion produced by an Evaluation.
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM13
Constraints / invariantsEvaluation Result is not AI Evaluation Report, Intelligence Effect, Assurance Finding, Risk Tier or Product qualification by default.

296.3 producesEvaluationResult​

FieldValue
Preferred labelproduces Evaluation Result
DefinitionRelates an Evaluation to an Evaluation Result produced by that assessment.
Semantic authorityGEAIRM
Subject kindsEvaluation
Object kindsEvaluation Result
Directionalitydirected
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M13
Source sliceM13

296.4 No universal PASS/FAIL vocabulary​

AIPS DPP Product Assurance currently uses:

PASS
PASS_WITH_CONDITIONS
FAIL
NOT_APPLICABLE

COVP Fitness for Purpose uses:

FIT
CONDITIONAL
UNFIT
INSUFFICIENT_EVIDENCE

Those are authoritative in their respective AIPS profiles. M13 does not generalize either list into a universal GEAIRM Evaluation Result vocabulary.

M13-R06 - Result-vocabulary restraint: GEAIRM MUST NOT impose AIPS Assurance result values or COVP Fitness statuses as universal Evaluation Result values.

M13-R07 - Evaluation-result non-collapse: Evaluation Result MUST remain distinct from AIPS Intelligence Effect, AI Product qualification, Effective Risk Tier and Assurance Finding unless explicitly mapped in a declared context.


297. AI Evaluation Report​

AIPS AIEvalReport remains an AI Asset.

M13 distinguishes the assessment activity from the report artifact:

Evaluation
assessment activity

Evaluation Result
finding produced by the assessment

AIEvalReport
AIPS AI Asset recording/presenting evaluation information

An Evaluation may produce multiple Results and those Results may be recorded in one or more Evaluation Reports.

Conversely, an AIEvalReport may aggregate results from multiple evaluation activities.

M13 does not broaden or redefine the AIPS AIEvalReport class.

297.1 recordedInEvaluationReport​

M13 introduces a generic relationship:

FieldValue
Preferred labelrecorded in AI Evaluation Report
DefinitionRelates an Evaluation Result to an AIPS AI Evaluation Report in which that result is recorded, summarized or referenced.
Semantic authorityGEAIRM relationship; report identity remains AIPS
Subject kindsEvaluation Result
Object kindsAIPS AIEvalReport
Directionalitydirected
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M13
Source sliceM13

M13-R08 - Evaluation/report distinction: AIPS AI Evaluation Report MUST remain distinct from the Evaluation activity and Evaluation Results it records.


298. Evidence as a Contextual Role​

M01 identified AI Evidence as a semantic-gap candidate but noted that AIPS already uses multiple PROV-based evidence specializations rather than one generic evidence superclass.

M13 resolves the gap by modelling Evidence as a contextual Role, not as a universal intrinsic superclass.

298.1 Canonical definition​

Evidence is a contextual Role played by a resource, record, observation, result, trace, report or other information-bearing entity when it is used to support, challenge, qualify or justify an Evaluation, Assurance Assessment, Risk Driver Assessment, Governance decision or other declared claim in a specified context.

The underlying resource retains its own identity.

Potential evidence resources include:

  • AI Qualification Evidence;
  • AI Primitive Evidence;
  • AI Evaluation Reports;
  • Evaluation Results;
  • metrics;
  • traces and logs;
  • audit records;
  • safety evidence;
  • red-team results;
  • COVP observations;
  • fitness assessments;
  • dependency/currentness records;
  • incident records;
  • provenance records;
  • human review records.

298.2 Evidence Role is not proof​

Evidence may be:

  • supportive;
  • contradictory;
  • incomplete;
  • stale;
  • scoped;
  • uncertain;
  • insufficient.

Therefore:

Evidence != proof
Evidence presence != correctness
Evidence presence != Assurance

298.3 Evidence Role vs AIPS ExplanationEvidence​

AIPS ExplanationEvidence is an Intelligence Effect value.

GEAIRM Evidence is a contextual Role.

An ExplanationEvidence occurrence or materialized artifact may play an Evidence Role in an Evaluation or Assurance context, but:

ExplanationEvidence != Evidence Role

298.4 Evidence Role vs M03 Evidence Reference​

M03 Evidence Reference is an L0 metamodel traceability construct used to support specification/model assertions.

M13 Evidence is an L1 enterprise-AI semantic Role.

They remain distinct.

M13-R09 - Evidence role resolution: GEAIRM MUST model generic Evidence as a contextual Role rather than create a universal superclass that replaces AIPS/PROV evidence identities.

M13-R10 - Evidence/proof distinction: Evidence MUST NOT be treated as proof, correctness, truth or sufficient assurance merely because it is present.

M13-R11 - Evidence-level distinction: M13 Evidence Role MUST remain distinct from M03 Evidence Reference and AIPS ExplanationEvidence.


299. Evidence Relationships​

299.1 usesEvidence​

FieldValue
Preferred labeluses Evidence
DefinitionRelates an Evaluation or Assurance Assessment to a resource playing the Evidence Role in that assessment context.
Semantic authorityGEAIRM
Subject kindsEvaluation or Assurance Assessment
Object kindsresource playing Evidence Role
Directionalitydirected
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M13
Source sliceM13

usesEvidence does not imply that the evidence is accepted, sufficient, authoritative, current or supportive.

299.2 Reuse of AIPS evidence references​

AIPS already provides scoped evidence-reference properties such as:

RiskDriverAssessment.evidenceRef
GovernanceObligationProfile.governanceEvidenceRef

M13 does not replace them with usesEvidence.

The generic relationship supports reference-model semantics; AIPS properties remain authoritative for AIPS Product risk/governance serialization.

M13-R12 - Evidence-property preservation: Generic usesEvidence MUST NOT replace or broaden AIPS evidence-reference properties in their governed Product scopes.


300. Assurance​

The word assurance can describe a process, body of evidence, outcome, organizational function or confidence judgment. AIPS already uses AI Product Assurance for Product-version trust evidence.

To avoid collision, GEAIRM does not mint a universal standalone Assurance entity. Instead M13 introduces the more precise Assurance Assessment and Assurance Finding concepts.

300.1 Assurance Assessment​

Assurance Assessment is a context-bounded, evidence-backed assessment of whether specified properties, claims, obligations or fitness expectations for a declared target are sufficiently supported within a stated assurance scope and conditions.

Assurance Assessment may use:

  • Evaluation Results;
  • AI Evaluation Reports;
  • runtime observations;
  • governance evidence;
  • safety evidence;
  • audit evidence;
  • fitness assessments;
  • qualification/currentness evidence;
  • human review;
  • other applicable evidence.

300.2 Element record​

FieldValue
Preferred labelAssurance Assessment
Element kindConcept / assessment activity
DefinitionA context-bounded, evidence-backed assessment of whether specified properties, claims, obligations or fitness expectations are sufficiently supported within a stated scope and conditions.
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM13
Constraints / invariantsAssurance Assessment is not AIPS AI Product Assurance record, Evaluation, Risk Tier, Governance Obligation, Control or approval authority.

300.3 assessesAssuranceOf​

FieldValue
Preferred labelassesses assurance of
DefinitionRelates an Assurance Assessment to the participant, property, claim or governed concern whose evidential support is being assessed in the declared scope.
Semantic authorityGEAIRM
Subject kindsAssurance Assessment
Object kindsapplicable participant/property/claim
Directionalitydirected
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M13
Source sliceM13

M13-R13 - No universal Assurance entity: GEAIRM 0.1 MUST NOT mint one overloaded Assurance superclass that conflates assurance process, evidence, result and AIPS Product-assurance records.

M13-R14 - Assurance/Evaluation distinction: Assurance Assessment MUST remain distinct from Evaluation; Evaluations may contribute evidence/results to Assurance without being identical to Assurance.


301. Assurance Finding​

301.1 Canonical definition​

Assurance Finding is a context-bounded conclusion produced by an Assurance Assessment regarding the sufficiency, insufficiency, conditionality or applicability of evidence for the declared assurance scope.

M13 does not impose one universal controlled vocabulary for Assurance Finding.

The AIPS DPP result vocabulary remains:

PASS
PASS_WITH_CONDITIONS
FAIL
NOT_APPLICABLE

for its Product Assurance representation.

Other assurance regimes may use other outcomes.

301.2 Element record​

FieldValue
Preferred labelAssurance Finding
Element kindConcept
DefinitionA context-bounded conclusion from an Assurance Assessment concerning evidential sufficiency or applicability for the declared scope.
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM13
Constraints / invariantsAssurance Finding is not Evaluation Result, Effective Risk Tier, AI Product qualification, Governance Obligation or authorization.

301.3 producesAssuranceFinding​

FieldValue
Preferred labelproduces Assurance Finding
DefinitionRelates an Assurance Assessment to an Assurance Finding produced for the declared assurance scope.
Semantic authorityGEAIRM
Subject kindsAssurance Assessment
Object kindsAssurance Finding
Directionalitydirected
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M13
Source sliceM13

M13-R15 - Assurance-finding vocabulary restraint: GEAIRM MUST NOT generalize the AIPS DPP PASS/PASS_WITH_CONDITIONS/FAIL/NOT_APPLICABLE vocabulary into a universal assurance-outcome taxonomy.

M13-R16 - Assurance-finding/risk distinction: Assurance Finding MUST NOT by itself determine AIPS Effective Risk Tier.


302. AIPS AI Product Assurance​

AIPS defines AI Product Assurance as evidence that the Product's governed intelligence proposition remains fit for its intended use. The AIPS DPP exposes Product assurance within a version-specific trust passport.

The AIPS DPP assurance representation currently supports scopes such as:

Product
IntelligenceCapability
AIPrimitive
Safety
Domain
Governance

and result values:

PASS
PASS_WITH_CONDITIONS
FAIL
NOT_APPLICABLE

M13 treats this as an AIPS Product-specific specialization/representation of assurance semantics.

GEAIRM generic Assurance Assessment may help explain how evidence is evaluated, but it does not replace the AIPS DPP assurance section or its conformance requirements.

M13-R17 - Product-assurance non-duplication: GEAIRM Assurance Assessment MUST NOT replace the mandatory AIPS AI Product Assurance representation for an AIPS-governed AI Product version.


303. Fitness for Purpose​

AIPS COVP defines Fitness for Purpose as a contextual assessment of whether a specific Product version is suitable for a specific intended use under stated conditions.

Its canonical conceptual statuses are:

FIT
CONDITIONAL
UNFIT
INSUFFICIENT_EVIDENCE

M13 reuses/references these semantics but does not generalize the COVP statuses to all Evaluation or Assurance.

The distinctions are:

Evaluation
assesses a target against criteria

Assurance Assessment
assesses sufficiency of evidence for declared assurance concerns

Fitness for Purpose
contextual Product-version suitability for a specific intended use

Effective Risk Tier
computed AIPS Product-version risk posture

A Product may be high risk and still be FIT for a narrowly governed intended use.

A Product may be low risk and be UNFIT for a particular intended use.

Likewise, a PASS assurance result does not imply FIT for every use.

M13-R18 - Fitness contextuality: COVP Fitness for Purpose MUST remain intended-use-specific and MUST NOT be treated as universal Product quality or universal assurance.

M13-R19 - Fitness/risk orthogonality: Fitness status MUST NOT directly determine AIPS R0-R4 Effective Risk Tier.


304. Risk Semantics and Effective Risk Tier​

M13 does not create a generic GEAIRM risk scale.

AIPS remains authoritative for AI Product-version Effective Risk Tier:

R0 - Minimal
R1 - Limited
R2 - Moderate / Material
R3 - High
R4 - Critical / Escalation-required

Every AIPS-governed AI Product version exposes exactly one computed Effective Risk Tier under the AIPS governance model.

The tier is evidence-backed rather than merely declared.

Conceptually AIPS computes the effective tier from applicable drivers including:

  • declared risk;
  • impact;
  • Agency / Autonomy;
  • Intelligence Effect;
  • Embodiment;
  • Topology;
  • domain sensitivity;
  • decision finality;
  • action authority;
  • impact scale;
  • reversibility;
  • qualification currentness;
  • dependency topology.

M13 does not reproduce the AIPS calculation as a GEAIRM formula.

304.1 Risk Tier is not Product Type​

M12 already established Product Type boundaries. M13 reinforces:

Risk Tier != AI Product Type
Risk Tier != Intelligence Capability
Risk Tier != Assurance Finding
Risk Tier != Fitness status

A Primary Product may be R4 and a Derived Product may be R1 under AIPS.

304.2 Risk Tier is not confidence​

R0-R4 expresses governed risk posture, not confidence in the model or probability that an output is correct.

304.3 Risk applies at the AIPS Product-version context​

GEAIRM MUST NOT assign AIPS R0-R4 directly to arbitrary AI Primitives, Agents, AI Executions, Tools or Compositions unless those resources participate in a governed AIPS Product context that explicitly carries the Product risk semantics.

M13-R20 - Risk-taxonomy non-duplication: GEAIRM MUST NOT create a second R0-R4 scale or redefine AIPS Effective Risk Tier.

M13-R21 - Risk/product-scope restraint: AIPS R0-R4 MUST NOT be directly generalized as an intrinsic classification of arbitrary non-Product GEAIRM participants.

M13-R22 - Risk-axis distinction: Effective Risk Tier MUST remain distinct from APT, capability, autonomy, Embodiment, Evaluation Result, Assurance Finding and Fitness status.


305. Risk Driver Assessment​

AIPS RiskDriverAssessment is an evidence-backed risk contribution from one orthogonal governance driver.

The current AIPS RiskDriverKind vocabulary includes:

DeclaredRisk
Impact
Agency
Effect
EmbodimentRisk
Topology
DomainSensitivity
DecisionFinality
ActionAuthority
ImpactScale
Reversibility
QualificationCurrentness
DependencyTopology

Each material driver assessment identifies:

  • driver kind;
  • contributed tier;
  • rationale;
  • evidence reference where applicable.

M13 reuses this structure without creating a GEAIRM RiskAssessment replacement.

An Evaluation Result or resource playing Evidence Role may be referenced by an AIPS Risk Driver Assessment where relevant, but the relationship must use the applicable AIPS governance semantics.

M13-R23 - Risk-driver authority preservation: GEAIRM MUST preserve AIPS Risk Driver Assessment semantics and MUST NOT replace it with a competing generic Product-risk record.

M13-R24 - Evidence/risk restraint: Evidence or Evaluation Results MUST NOT themselves determine a Risk Tier absent the governed AIPS risk computation and applicable driver rules.


306. AIPS Hard Risk Floors Remain External​

AIPS establishes Product-governance hard floors, including:

AR = AutonomousAction
-> Effective Risk Tier >= R2

AR = AutonomousAction
AND IE includes PhysicalAction
-> Effective Risk Tier >= R3

EM = ClosedLoopAutonomous
AND IE includes PhysicalAction
-> Effective Risk Tier >= R3

AR = AutonomousAction
AND Topology Scope = EnterpriseOrchestrated
-> Effective Risk Tier = R4

AIPS also requires ActionAuthorization when:

AR in {ActWithApproval, AutonomousAction}

M13 references these rules only to preserve semantic alignment. It does not restate them as GEAIRM-owned risk rules.

M13-R25 - AIPS hard-floor authority: Product-level hard risk floors and mandatory obligation triggers remain AIPS-owned and MUST NOT be redefined by GEAIRM.


307. Governance Obligation Profile​

AIPS requires exactly one current GovernanceObligationProfile for the evaluated AI Product version.

The profile records:

  • the Effective Risk Tier from which baseline obligations were derived;
  • governance driver dimensions;
  • resulting governance obligations;
  • computation timestamp/method;
  • supporting evidence references where applicable.

AIPS governance stacking is additive:

GovernanceObligations =
RiskTierObligations
UNION OntologyDimensionObligations
UNION ImpactObligations
UNION TopologyObligations
UNION QualificationCurrentnessObligations
UNION DependencyCurrentnessObligations

M13 does not turn this into a generic GEAIRM governance-calculation algorithm for non-Product participants.

307.1 Current AIPS Governance Obligation vocabulary​

The current AIPS core includes:

ProductQualificationCurrentness
LineageProvenance
RuntimeObservability
ChangeQualityMonitoring
HumanReview
HumanApproval
InterventionOverride
ImmutableAudit
ExplainabilityTransparency
ActionAuthorization
SafetyCase
EmergencyStopSafeState
IndependentValidation
RedTeamEvaluation
CrossDomainCoordinationReview
ExecutiveEscalation
IncidentResponse
DependencyCurrentness
AbuseMisuseControls

M13 reuses those terms where an AIPS Product-version Governance Obligation Profile is involved.

M13-R26 - Governance-profile non-duplication: GEAIRM MUST NOT create a second Governance Obligation Profile or shadow AIPS Governance Obligation vocabulary.

M13-R27 - Obligation stacking preservation: GEAIRM MUST NOT suppress or reinterpret an AIPS governance obligation merely because another ontology/risk dimension appears lower risk.


308. Risk Tier vs Governance Obligation​

Risk and governance obligation are distinct:

Effective Risk Tier
computed risk posture

Governance Obligation Profile
controls/requirements that must apply
based on tier plus ontology, impact,
topology and currentness drivers

A lower risk tier does not imply absence of all governance requirements because:

  • baseline Product controls may still apply;
  • ontology dimensions may add obligations;
  • domain/regulatory policy may add obligations;
  • topology/currentness may add obligations.

Conversely, presence of a governance obligation does not by itself prove a particular Risk Tier unless AIPS explicitly establishes the trigger.

M13-R28 - Risk/obligation distinction: Effective Risk Tier and Governance Obligation Profile MUST remain distinct and MUST NOT be inferred from one another except through authoritative AIPS rules.


309. Governance Obligation vs Governance Control​

AIPS GovernanceObligation represents what governance requires for an AI Product version.

AIPDS separately defines GovernanceControlBinding as an operational binding from a Governance Obligation Profile obligation to an enforceable deployment control.

This yields a critical distinction:

Governance Obligation
what must be satisfied

Governance Control
mechanism/process used to satisfy or enforce it

Governance Control Binding
deployment-specific binding connecting
obligation to enforceable control

A single Governance Obligation may be realized by multiple controls.

One control may contribute to multiple obligations where explicitly governed.

M13 does not define the implementation topology of those controls.

309.1 Control lexical collision​

M13 preserves the following separation:

AIPS Control
Intelligence Capability

Agent ControlMode
Agent oversight facet

Governance Control
governance/assurance mechanism

Control Loop
environment interaction pattern

These are not equivalent.

M13-R29 - Obligation/control distinction: Governance Obligation MUST remain distinct from the governance Control that realizes or enforces it.

M13-R30 - Control collision discipline: AIPS Control Intelligence Capability, Agent ControlMode, GEAIRM Control Loop and governance Control MUST remain semantically distinct.

M13-R31 - AIPDS realization boundary: AIPDS Governance Control Binding MUST remain an operational/deployment realization construct and MUST NOT be redefined as a generic GEAIRM semantic obligation.


310. Policy, Rule, Instruction, Obligation and Control​

M07 intentionally deferred full Policy/Control semantics. M13 now resolves the boundary without creating a universal policy ontology.

GEAIRM treats Policy as an externally governed normative statement under the applicable enterprise, legal, regulatory, security, safety or domain authority.

GEAIRM does not mint one universal Policy class in 0.1 because policy semantics vary materially across governing frameworks.

The semantic distinctions are:

Policy
externally governed normative statement

Rule
declarative/logical/conditional statement
established in M07

Instruction
directive intended to influence behaviour
established in M07

Governance Obligation
AIPS Product-version requirement

Governance Control
mechanism/process realizing or enforcing governance

Evaluation Criterion
property/threshold/question used in assessment

Possible contextual mappings include:

Policy
may create or inform
Governance Obligation

Policy
may be operationalized by
Rule

Governance Obligation
may be realized by
Governance Control

Policy / Obligation
may inform
Evaluation Criterion

None is semantic equivalence.

M13-R32 - Policy externality: GEAIRM 0.1 MUST NOT create a universal Policy ontology that overrides applicable enterprise/regulatory governance authorities.

M13-R33 - Governance-term separation: Policy, Rule, Instruction, Governance Obligation, Governance Control and Evaluation Criterion MUST remain distinct unless an explicit governed mapping is established.


311. Evaluation, Assurance and Governance Are Not the Same Thing​

M13 establishes a useful three-stage conceptual distinction:

Evaluation
What did we assess and what did we observe/find?

Assurance Assessment
Is the body of evidence sufficient for the declared
assurance concern and scope?

Governance
What requirements, constraints, responsibilities
and controls apply?

A successful Evaluation does not automatically imply adequate Assurance.

An adequate Assurance Finding does not eliminate Governance Obligations.

A Governance Obligation may require an Evaluation or Assurance activity, but it does not become that activity.

For example:

RedTeamEvaluation
AIPS Governance Obligation

may require

Evaluation activity
using red-team methods

producing

Evaluation Results / AIEvalReport

contributing to

Assurance Assessment

The terms remain distinct.

M13-R34 - Evaluation/assurance/governance separation: Evaluation, Assurance Assessment and Governance Obligation MUST remain distinct semantic concerns.


312. Assurance Is Mechanism-Neutral​

AIPS states that AI Product Assurance is mechanism-neutral and can evaluate learned, symbolic, agentic, hybrid or Physical AI Products.

GEAIRM extends only the generic semantic principle:

Assurance evidence and Evaluation methods MUST match the actual target semantics and constitution rather than assuming every AI participant is a learned model.

Therefore a model card is relevant when a learned model is materially involved, but it is not a universal assurance requirement for symbolic, knowledge-reasoning, planning or model-free AI.

Likewise, "model drift" is not a universal change/quality concept.

Applicable evidence may instead include:

  • rule/knowledge change analysis;
  • logical consistency;
  • environment coverage;
  • tool/dependency behaviour;
  • plan validity;
  • authorization traces;
  • physical safety evidence;
  • runtime effect evidence;
  • model evaluation where learned models apply.

M13-R35 - Mechanism-neutral assurance: GEAIRM Evaluation and Assurance semantics MUST remain mechanism-neutral and MUST NOT impose learned-model-specific evidence on all enterprise AI.


313. Assurance vs Risk​

Assurance and risk answer different questions.

Assurance
Is there sufficient evidence supporting the declared
properties/fitness/obligations within scope?

Risk
What governed risk posture follows from applicable
risk drivers and impact context?

A high-risk Product may have strong, current assurance evidence.

A low-risk Product may have weak or stale assurance evidence.

Therefore:

PASS assurance != R0/R1
FAIL assurance != R4 automatically
high assurance != low risk
high risk != failed assurance

A material assurance failure may trigger risk re-evaluation, governance escalation or lifecycle action under applicable AIPS rules, but M13 does not define a direct one-to-one mapping.

M13-R36 - Assurance/risk orthogonality: Assurance outcome MUST NOT be used as a direct substitute for Effective Risk Tier, and Effective Risk Tier MUST NOT be treated as an assurance outcome.


314. Runtime Evidence and Continuous Assurance​

AIPS requires declared ontology/governance characteristics to remain consistent with observed runtime behaviour and describes Continuous Assurance as comparing runtime evidence with declared Product characteristics and governance assumptions.

M13 adopts the generic semantic pattern:

Declared semantics / obligations
|
v
Runtime observations / evidence
|
v
Evaluation
|
v
Assurance Assessment
|
material mismatch?
/ \
no yes
|
v
reassessment / escalation /
lifecycle or governance action

This pattern does not mean that runtime evidence directly rewrites ontology.

For example:

observed wider topology
!= automatic Topology Scope mutation

observed physical action
!= automatic Embodiment mutation

adverse fitness evidence
!= automatic APT mutation

Instead, material mismatch triggers governed reassessment and, where necessary, a new Product/version or lifecycle decision.

Detailed currentness states and change routing are deferred to M14 and AIPS lifecycle specifications.

M13-R37 - Runtime-evidence restraint: Runtime evidence MUST NOT directly mutate Product Family, APT, Embodiment, Agency/Autonomy, Topology Scope or Risk Tier without the applicable governed reassessment.

M13-R38 - Continuous-assurance/lifecycle boundary: M13 MUST NOT duplicate the detailed currentness, supersession and change-routing semantics reserved for AIPS lifecycle specifications and M14.


315. Observability Is Not Assurance​

Runtime observability provides signals, telemetry, traces, logs, rationales and other records that may serve as Evidence.

But:

Observability != Evaluation
Observability != Assurance
Observability != Governance compliance

A highly observable system may still fail an Evaluation or lack sufficient evidence for Assurance.

A system may satisfy an assurance requirement using evidence that is not runtime telemetry.

AIPS RuntimeObservability is also a Governance Obligation term; that obligation is not identical to the telemetry itself.

M13-R39 - Observability/assurance distinction: Observability signals and RuntimeObservability obligations MUST NOT be treated as equivalent to Evaluation, Assurance Finding or governance compliance.


316. Governance Evidence Does Not Prove Governance Compliance​

A Governance Obligation Profile may reference governance evidence. Presence of evidence does not necessarily prove that the obligation has been fully and correctly realized.

For example:

  • an audit log existing does not prove every required event is captured;
  • a safety case document existing does not prove the deployed system satisfies it;
  • an approval record existing does not prove the approved action remained within scope;
  • a red-team report existing does not prove every identified issue is resolved.

Evaluation and Assurance may be required to assess the adequacy of that evidence.

M13-R40 - Governance-evidence restraint: Presence of governance evidence MUST NOT by itself establish that a Governance Obligation is satisfied.


317. Governance Obligations Do Not Transfer Through Dependency Automatically​

M11 established that dependency and composition do not automatically propagate policy, risk or governance semantics.

M13 preserves that principle.

If Product A depends on Product B:

B has Governance Obligation X

does not automatically imply

A has Governance Obligation X

A may independently acquire:

  • its own obligation X;
  • a dependency-currentness obligation;
  • a provenance/lineage obligation;
  • an integration-specific obligation;
  • no equivalent obligation at all;

depending on AIPS governance computation and applicable external policy.

Likewise, a downstream Product does not inherit the upstream Risk Tier by graph structure alone.

M13-R41 - Governance non-inheritance: Governance Obligations and Effective Risk Tier MUST NOT automatically propagate across dependency, Composition, Federation or Orchestration relationships.


318. Authorization and Action Governance​

M10 introduced GEAIRM Authorization and preserved AIPS ActionAuthorization as a Governance Obligation.

These are distinct:

Authorization
context-bounded grant / approval / permission
for a specific occurrence

ActionAuthorization
AIPS Governance Obligation requiring
appropriate action-authorization governance

A Governance Obligation requiring ActionAuthorization may be realized through one or more authorization mechanisms and may result in concrete M10 Authorization records/occurrences.

The obligation itself is not the authorization instance.

Likewise:

HumanApproval obligation != concrete Authorization
InterventionOverride obligation != actual intervention occurrence
EmergencyStopSafeState obligation != a specific emergency-stop event

M13-R42 - Obligation/occurrence distinction: AIPS governance obligations such as ActionAuthorization, HumanApproval, InterventionOverride and EmergencyStopSafeState MUST remain distinct from the concrete authorization, approval or intervention occurrences that satisfy them.


319. Evaluation and Assurance of Non-Product AI​

GEAIRM applies more broadly than AIPS Product qualification. Therefore generic Evaluation and Assurance Assessment may apply to non-Product AI participants such as:

  • an AI Primitive under development;
  • an Agent;
  • a Tool;
  • an AI Execution;
  • a Composition;
  • an Environment interaction;
  • an experimental mechanism.

However, using GEAIRM Evaluation or Assurance semantics on such a participant does not:

  • make it a Product;
  • make it an AI Product;
  • assign APT;
  • assign AIPS R0-R4;
  • create a Governance Obligation Profile.

AIPS Product governance applies when the applicable BPS/AIPS Product boundary and qualification are established.

M13-R43 - Non-Product assessment boundary: Evaluation or Assurance of a non-Product AI participant MUST NOT by itself assign Product identity, APT, AIPS Effective Risk Tier or Governance Obligation Profile.


320. Canonical M13 Reference View​

                           EVALUATION

+------------------+
| Evaluation |
| GEAIRM-owned |
+----+--------+----+
| |
evaluatesTarget | usesEvaluationCriterion
| |
v v
Target Evaluation Criterion
|
|
+---- usesEvidence ----> resource playing
Evidence Role
|
+---- producesEvaluationResult
|
v
Evaluation Result
|
+---- may be recorded in
v
AIPS AIEvalReport


ASSURANCE

+----------------------+
| Assurance Assessment |
| GEAIRM-owned |
+----+------------+----+
| |
assessesAssuranceOf +---- usesEvidence ---->
|
v
Target
|
v
Assurance Finding

AIPS AI Product Assurance remains the Product-version trust representation.
COVP Fitness for Purpose remains intended-use-specific Product evidence.


RISK & GOVERNANCE

AI Product Version
|
+---- Effective Risk Tier ----> R0 | R1 | R2 | R3 | R4
| |
| +---- has Risk Driver Assessment(s)
|
+---- Governance Obligation Profile
|
+---- Governance Obligation(s)
|
| operationalized in AIPDS
v
Governance Control Binding
|
v
enforceable control

Policy / regulatory authority
may constrain risk/governance/evaluation semantics
but is not replaced by GEAIRM.

Evaluation Result / Evidence
may support Risk Driver Assessment or Assurance,
but does not independently set Risk Tier.

The view is semantic. It does not prescribe evaluation services, model-monitoring platforms, policy engines, control planes, risk engines, evidence stores or assurance architecture.


321. Illustrative Semantic Examples​

321.1 Learned-model evaluation supporting Product assurance​

Evaluation E1
evaluatesTarget FraudModel-v4
usesEvaluationCriterion AccuracyThreshold
usesEvaluationCriterion CalibrationCriterion
producesEvaluationResult ER1

ER1
recordedInEvaluationReport EvalReport-4

Assurance Assessment AA1
assessesAssuranceOf FraudProduct-v4
usesEvidence EvalReport-4
producesAssuranceFinding AF1

The Evaluation Report may contribute to AIPS Product Assurance, but it is not itself the assurance conclusion.

321.2 Symbolic reasoner assurance without a Model Card​

Evaluation E2
evaluatesTarget PolicyReasoner-v2
usesEvaluationCriterion LogicalConsistency
usesEvaluationCriterion RuleCoverage

Assurance Assessment AA2
usesEvidence ReasoningEvalReport-v2

Nothing requires a Model Card merely because the target is AI.

321.3 High risk with successful assurance​

AI Product Version P3
EffectiveRiskTier = R3 [AIPS]

AIPS Product Assurance
result = PASS

No inference changes R3 to R1 or R0.

321.4 Fitness is contextual​

Product Version P4
Fitness for IntendedUse-A = FIT
Fitness for IntendedUse-B = INSUFFICIENT_EVIDENCE

There is no contradiction because Fitness is intended-use-specific.

321.5 Governance obligation realized through deployment control​

Governance Obligation Profile GOP5
hasGovernanceObligation ActionAuthorization

AIPDS Deployment D5
governanceControlBinding GCB5

GCB5
boundGovernanceObligation ActionAuthorization
controlRef Authorization-Control-5

The Governance Obligation is not the control implementation.

321.6 Runtime mismatch triggers reassessment​

Declared Product Topology Scope
DomainCoupled

Observed runtime evidence
indicates cross-domain propagation

M13 does not mutate the Product's Topology Scope from telemetry alone. The mismatch triggers governed re-evaluation under applicable AIPS lifecycle/governance semantics.

321.7 Evaluation of a non-Product Agent​

Agent A7
evaluated for tool-authority compliance

Evaluation E7
evaluatesTarget A7

No Product, AI Product, APT, R0-R4 or Governance Obligation Profile is inferred.


322. M13 Canonical Element Inventory​

ElementM03 kindAuthorityDispositionM13 status
EvaluationConcept / assessment activityGEAIRMOWNEDcanonical
Evaluation TargetRoleGEAIRMOWNED as contextual rolecanonical
Evaluation CriterionConceptGEAIRMOWNEDcanonical
Evaluation ResultConceptGEAIRMOWNEDcanonical
evaluatesTargetRelationship TypeGEAIRMOWNEDcanonical
usesEvaluationCriterionRelationship TypeGEAIRMOWNEDcanonical
producesEvaluationResultRelationship TypeGEAIRMOWNEDcanonical
AI Evaluation ReportAI AssetAIPSREUSEDcanonical external
recordedInEvaluationReportRelationship TypeGEAIRMOWNEDcanonical
EvidenceRoleGEAIRMOWNED as contextual rolecanonical
usesEvidenceRelationship TypeGEAIRMOWNEDcanonical
AI Qualification EvidenceEvidence entityAIPSREUSEDcanonical external
AI Primitive EvidenceEvidence entityAIPSREUSEDcanonical external
ExplanationEvidenceIntelligence Effect valueAIPSREUSEDcanonical external
Assuranceeditorial umbrellaGEAIRM editorial usagenot canonical standalone entityresolved
Assurance AssessmentConcept / assessment activityGEAIRMOWNEDcanonical
Assurance FindingConceptGEAIRMOWNEDcanonical
assessesAssuranceOfRelationship TypeGEAIRMOWNEDcanonical
producesAssuranceFindingRelationship TypeGEAIRMOWNEDcanonical
AI Product AssuranceProduct trust semanticsAIPS DPPREUSED / REFERENCEDcanonical external
Fitness for PurposeProduct consumption semanticsAIPS COVPREUSED / REFERENCEDcanonical external
Effective Risk TierConcept / computed risk postureAIPSREUSEDcanonical external
Risk Tier LevelClassificationAIPSREUSEDcanonical external
R0 / R1 / R2 / R3 / R4Classification valuesAIPSREUSEDcanonical external
Risk Driver AssessmentConceptAIPSREUSEDcanonical external
Risk Driver KindClassificationAIPSREUSEDcanonical external
Governance ObligationConcept/classification familyAIPSREUSEDcanonical external
Governance Obligation ProfileConceptAIPSREUSEDcanonical external
Policyexternal governance conceptapplicable external authorityREFERENCEDno GEAIRM shadow class
Governance Controlexternal/realization conceptgovernance framework / AIPDS / GEAIRAREFERENCED / RESERVEDno universal GEAIRM class
Governance Control Bindingdeployment bindingAIPDSREUSEDcanonical external at deployment scope

No executable GEAIRM identifiers are minted in M13. Stable machine identifiers remain deferred to the A1-A8 formalization programme after M17.


323. M13 Hard Distinctions​

The following distinctions are normative for subsequent slices:

Evaluation != AI Evaluation Report
Evaluation != Assurance Assessment
Evaluation != Risk Tier
Evaluation != Governance Obligation

Evaluation Criterion != Evaluation Result
Evaluation Criterion != Policy
Evaluation Criterion != Governance Obligation

Evaluation Result != AI Evaluation Report
Evaluation Result != Intelligence Effect automatically
Evaluation Result != Assurance Finding
Evaluation Result != Risk Tier

Evidence Role != intrinsic evidence superclass
Evidence != proof
Evidence != correctness
Evidence != Assurance
Evidence Role != ExplanationEvidence
Evidence Role != M03 Evidence Reference

Assurance Assessment != AI Product Assurance record
Assurance Assessment != Evaluation
Assurance Finding != Effective Risk Tier
Assurance Finding != Fitness status
PASS assurance != low risk

Fitness for Purpose != universal Product quality
Fitness status != Effective Risk Tier

Effective Risk Tier != APT
Effective Risk Tier != confidence
Effective Risk Tier != Assurance Finding

Governance Obligation != Governance Control
Governance Obligation != concrete authorization/intervention event
Governance Obligation Profile != Risk Tier

AIPS Control Intelligence Capability
!= Agent ControlMode
!= GEAIRM Control Loop
!= governance Control

Policy != Rule
Policy != Instruction
Policy != Governance Obligation
Policy != Governance Control

runtime evidence != ontology mutation
observability != assurance
governance evidence != proof of compliance

dependency/composition != governance inheritance

As in prior slices, != denotes semantic non-equivalence rather than formal OWL disjointness unless later executable formalization explicitly authorizes stronger axioms.


324. M13 Normative Rules​

RuleRequirement
M13-R01GEAIRM MUST preserve AIPS authority for Effective Risk Tier, R0-R4, Risk Driver Assessment, Governance Obligation and Governance Obligation Profile.
M13-R02GEAIRM MUST preserve AIPS authority for AI Product Assurance and COVP Fitness for Purpose in Product-version contexts.
M13-R03Conducting Evaluation MUST NOT by itself establish Artificial Intelligence, AI Primitive or AI Product identity for the evaluator.
M13-R04Playing the Evaluation Target Role MUST NOT alter Product Family, Agent identity, AI Primitive identity, risk classification or governance ownership.
M13-R05Evaluation Criterion MUST remain distinct from Policy, Rule, Governance Obligation and Governance Control absent explicit mapping.
M13-R06GEAIRM MUST NOT impose AIPS assurance-result values or COVP Fitness statuses as universal Evaluation Result values.
M13-R07Evaluation Result MUST remain distinct from Intelligence Effect, AI Product qualification, Effective Risk Tier and Assurance Finding unless explicitly mapped.
M13-R08AIPS AI Evaluation Report MUST remain distinct from Evaluation activity and Evaluation Results.
M13-R09GEAIRM MUST model generic Evidence as a contextual Role rather than replace AIPS/PROV evidence identities with a universal superclass.
M13-R10Evidence MUST NOT be treated as proof, correctness, truth or sufficient assurance merely because it is present.
M13-R11M13 Evidence Role MUST remain distinct from M03 Evidence Reference and AIPS ExplanationEvidence.
M13-R12Generic usesEvidence MUST NOT replace or broaden AIPS evidence-reference properties within governed Product risk/governance structures.
M13-R13GEAIRM 0.1 MUST NOT mint one overloaded Assurance superclass conflating process, evidence, result and Product-assurance records.
M13-R14Assurance Assessment MUST remain distinct from Evaluation; Evaluation may contribute to but does not equal Assurance.
M13-R15GEAIRM MUST NOT generalize the AIPS DPP assurance result vocabulary into a universal assurance-outcome taxonomy.
M13-R16Assurance Finding MUST NOT by itself determine AIPS Effective Risk Tier.
M13-R17GEAIRM Assurance Assessment MUST NOT replace mandatory AIPS AI Product Assurance representation for an AIPS-governed Product version.
M13-R18COVP Fitness for Purpose MUST remain intended-use-specific and MUST NOT be treated as universal Product quality or universal assurance.
M13-R19Fitness status MUST NOT directly determine AIPS R0-R4 Effective Risk Tier.
M13-R20GEAIRM MUST NOT create a second R0-R4 scale or redefine AIPS Effective Risk Tier.
M13-R21AIPS R0-R4 MUST NOT be generalized as intrinsic classification of arbitrary non-Product GEAIRM participants.
M13-R22Effective Risk Tier MUST remain distinct from APT, capability, autonomy, Embodiment, Evaluation Result, Assurance Finding and Fitness status.
M13-R23GEAIRM MUST preserve AIPS Risk Driver Assessment semantics and MUST NOT replace them with a competing Product-risk record.
M13-R24Evidence or Evaluation Results MUST NOT themselves determine Risk Tier absent governed AIPS risk computation.
M13-R25AIPS hard risk floors and mandatory Product obligation triggers remain AIPS-owned and MUST NOT be redefined by GEAIRM.
M13-R26GEAIRM MUST NOT create a second Governance Obligation Profile or shadow AIPS Governance Obligation vocabulary.
M13-R27GEAIRM MUST NOT suppress an AIPS Governance Obligation merely because another driver appears lower risk.
M13-R28Effective Risk Tier and Governance Obligation Profile MUST remain distinct except through authoritative AIPS rules.
M13-R29Governance Obligation MUST remain distinct from the Governance Control that realizes or enforces it.
M13-R30AIPS Control Intelligence Capability, Agent ControlMode, GEAIRM Control Loop and governance Control MUST remain distinct.
M13-R31AIPDS Governance Control Binding MUST remain a deployment realization construct and MUST NOT be redefined as a generic GEAIRM obligation.
M13-R32GEAIRM 0.1 MUST NOT create a universal Policy ontology overriding applicable enterprise/regulatory governance authorities.
M13-R33Policy, Rule, Instruction, Governance Obligation, Governance Control and Evaluation Criterion MUST remain distinct unless explicitly mapped.
M13-R34Evaluation, Assurance Assessment and Governance Obligation MUST remain distinct semantic concerns.
M13-R35Evaluation and Assurance semantics MUST remain mechanism-neutral and MUST NOT impose learned-model-specific evidence on all AI.
M13-R36Assurance outcome MUST NOT substitute for Effective Risk Tier, and Effective Risk Tier MUST NOT be treated as an assurance outcome.
M13-R37Runtime evidence MUST NOT directly mutate Product Family, APT, Embodiment, Agency/Autonomy, Topology Scope or Risk Tier without governed reassessment.
M13-R38M13 MUST NOT duplicate detailed currentness, supersession and change-routing semantics reserved for AIPS lifecycle specifications and M14.
M13-R39Observability signals and RuntimeObservability obligations MUST NOT be treated as equivalent to Evaluation, Assurance Finding or governance compliance.
M13-R40Presence of governance evidence MUST NOT by itself establish that a Governance Obligation is satisfied.
M13-R41Governance Obligations and Effective Risk Tier MUST NOT automatically propagate across dependency, Composition, Federation or Orchestration.
M13-R42Governance obligations such as ActionAuthorization, HumanApproval, InterventionOverride and EmergencyStopSafeState MUST remain distinct from concrete occurrences that satisfy them.
M13-R43Evaluation or Assurance of a non-Product AI participant MUST NOT by itself assign Product identity, APT, AIPS Effective Risk Tier or Governance Obligation Profile.
M13-R44M13 relationships MUST follow M03 conservative inference defaults unless stronger semantics are explicitly established.
M13-R45M13 MUST NOT prescribe evaluation platforms, risk engines, policy engines, assurance services, control planes or governance workflow architecture.
M13-R46M13 semantic non-equivalence MUST NOT be strengthened into formal OWL disjointness unless later executable formalization explicitly authorizes it.

325. M13 Exit Criteria​

GEAIRM-M13 is complete when:

  • generic Evaluation is defined as an assessment activity rather than an AI Evaluation Report;
  • Evaluation Target is a contextual Role rather than a new intrinsic hierarchy;
  • Evaluation Criterion and Evaluation Result are explicitly represented;
  • no universal Evaluation Result vocabulary is imposed;
  • AIPS AI Evaluation Report remains an AIPS AI Asset;
  • the M01 AI Evidence gap is resolved through a contextual Evidence Role rather than a replacement evidence superclass;
  • Evidence is separated from proof, correctness and Assurance;
  • Evidence Role is separated from AIPS ExplanationEvidence and M03 Evidence Reference;
  • Assurance is resolved through Assurance Assessment and Assurance Finding rather than an overloaded universal class;
  • AIPS AI Product Assurance remains authoritative for Product trust;
  • COVP Fitness for Purpose remains contextual and intended-use-specific;
  • AIPS Effective Risk Tier R0-R4 remains authoritative and Product-scoped;
  • AIPS Risk Driver Assessment remains authoritative;
  • AIPS hard risk floors remain external rather than duplicated;
  • Governance Obligation Profile and Governance Obligation vocabulary remain AIPS-owned;
  • Risk Tier is separated from Governance Obligation Profile;
  • Governance Obligation is separated from the Control that realizes it;
  • AIPDS Governance Control Binding remains deployment-specific;
  • Policy remains externally governed rather than replaced by a universal GEAIRM policy ontology;
  • Policy, Rule, Instruction, Obligation, Control and Evaluation Criterion remain distinct;
  • Evaluation, Assurance and Governance remain separate concerns;
  • assurance remains mechanism-neutral;
  • Assurance outcome remains orthogonal to Risk Tier;
  • runtime evidence triggers governed reassessment rather than direct ontology mutation;
  • observability remains evidence-producing rather than equivalent to Assurance;
  • governance evidence does not by itself prove compliance;
  • governance and risk do not automatically propagate through dependency/composition;
  • M14 retains detailed currentness, lifecycle, identity and provenance scope;
  • no evaluation/risk/policy/assurance architecture is prescribed;
  • M01-M12 authority, abstraction and conservative-inference rules remain intact.

M13 conclusion: these criteria are satisfied by this slice. GEAIRM now has a generic, mechanism-neutral semantic model for Evaluation, Evidence and Assurance that can integrate with enterprise AI participants broadly while preserving AIPS Product-specific Effective Risk Tier, Risk Driver Assessment, Governance Obligation Profile, AI Product Assurance and AIPDS control-realization authority.


326. Forward Work​

The next slice is:

GEAIRM-M14 - Identity, Version, Lifecycle, Currentness & Provenance

M14 will define how identity, version, dependency binding, lifecycle state, qualification/governance/assurance currentness, provenance and semantic change interact. It will preserve historical truth, distinguish supersession from mutation, and reconcile GEAIRM reference-model participants with the version-specific AIPS Product lifecycle without turning every AI resource into a Product version.


327. GEAIRM-M14 - Identity, Version, Lifecycle, Currentness & Provenance​

327.1 Purpose of this slice​

M13 established Evaluation, Evidence, Assurance, Risk and Governance semantics while deliberately deferring detailed currentness and lifecycle semantics. M14 now establishes how enterprise-AI participants preserve identity through change, how versions represent historical semantic states, how lifecycle and currentness are kept distinct, and how provenance and lineage explain origin and change without rewriting history.

M14 answers ten questions:

  1. How does continuing identity differ from a version-specific semantic state?
  2. How does GEAIRM integrate the AIPS ProductIdentity / AIProductVersion model without turning every AI participant into a Product version?
  3. What generic version semantics are required for AI Primitives, Agents, Knowledge Resources, Tools, Evaluations and other non-Product participants?
  4. How does generic Lifecycle State differ from AIPS LifecycleStage, PDEP process stage, Marketplace listing state and Continuous Assurance status?
  5. What is Currentness, and why is it a time- and concern-qualified assessment rather than an intrinsic identity property?
  6. How do AIPS qualification, dependency, governance and assurance currentness relate without becoming one universal currentness taxonomy?
  7. How do supersession, deprecation, retirement and historical closure preserve historical semantic truth?
  8. How should GEAIRM use provenance without creating a competing provenance ontology alongside W3C PROV and AIPS provenance semantics?
  9. How do provenance, lineage and dependency differ?
  10. How should material change trigger reassessment and downstream impact without silently mutating existing historical versions?

M14 does not prescribe a version-control system, registry implementation, release pipeline, graph database, event-sourcing platform, artifact store, lineage engine or lifecycle workflow. Those are architecture and implementation concerns reserved for GEAIRA, AIPS tooling and enterprise platforms.


328. Source Semantics and Authority Resolution​

M14 is governed by M01 authority rules and the historical/non-inheritance principles established in M03-M13.

The current AIPS core establishes:

aips:ProductIdentity
continuing logical Product identity across immutable Product versions

aips:AIProductVersion
immutable governed version with evidence-backed AI Product Family qualification

aips:hasVersion
ProductIdentity -> AIProductVersion

AIPS further establishes:

QualificationStatus
CURRENT
STALE
REVALIDATION_REQUIRED
INVALID

LifecycleStage
Experimental
Beta
Production
Deprecated
Retired

The AIPS Product lifecycle documentation states that Product identity may continue while Product versions remain immutable and carry their own qualification, APT, governance and lineage. It also states that later versions must not rewrite the historical ontology of earlier versions.

PDEP defines TrustCurrentness with separate currentness dimensions for:

qualification
dependencies
governance
assurance

using the AIPS QualificationStatus vocabulary in the current PDEP 0.1 machine model.

AIPS Marketplace separately defines listing states:

Active
Restricted
Suspended
Historical
Retired

and COVP defines its own currentness semantics for consumption evidence. These are not treated as one vocabulary.

AIPS uses W3C PROV concepts and properties throughout the current machine-readable baseline, including prov:Entity and prov:wasDerivedFrom. M14 therefore treats PROV as the generic external provenance foundation rather than inventing a competing provenance ontology.

M14 adopts the following authority treatment:

ConcernSemantic authorityM14 treatment
generic identity disciplineGEAIRM modelling semantics + applicable external identity authorityintegrated; no universal identity replacement
Product IdentityAIPS/BPS Product authorityREUSED
AI Product VersionAIPSREUSED
hasVersion for AIPS Product identityAIPSREUSED
generic Versioned ResourceGEAIRMOWNED as contextual Role
generic Version IdentifierGEAIRMOWNED as identifying value/record concept
generic Lifecycle StateGEAIRMOWNED as M03 State
AIPS Lifecycle Stage and valuesAIPSREUSED
PDEP process stageAIPS PDEPREFERENCED; not Lifecycle State
generic Currentness AssessmentGEAIRMOWNED
AIPS QualificationStatusAIPSREUSED in AIPS scopes
PDEP Trust CurrentnessAIPS PDEPREUSED / REFERENCED
Marketplace listing stateAIPS MarketplaceREFERENCED
COVP currentnessAIPS COVPREFERENCED
SupersessionGEAIRMOWNED as context-bounded relationship semantics
Historical ClosureAIPS lifecycle semanticsREUSED / REFERENCED
generic provenanceW3C PROVEXTERNAL-AUTHORITY / REUSED
AIPS Product lineage/provenanceAIPSREUSED / REFERENCED
detailed deployment/version rolloutAIPDS / GEAIRA / implementationRESERVED

M14-R01 - AIPS identity authority: GEAIRM MUST preserve AIPS authority for ProductIdentity, AIProductVersion and hasVersion in AIPS Product semantics.

M14-R02 - Provenance authority: GEAIRM MUST reuse or reference W3C PROV and applicable AIPS provenance semantics rather than create a competing generic provenance ontology.


329. Identity Discipline​

M03 established that identity is not the same thing as a label. M14 strengthens that rule across time.

For any governed participant, identity answers:

Which continuing thing is this?

Version answers:

Which governed semantic state or revision of that thing is being described?

Classification answers:

What is true of that thing or version under an applicable semantic authority?

These questions must remain separate.

A resource may preserve continuing identity even while a later version changes:

  • implementation mechanism;
  • AI Primitive dependency;
  • Intelligence Capability;
  • APT;
  • Product Family where the applicable Product identity model permits continuity;
  • risk/governance posture;
  • lifecycle/currentness state.

Identity continuity therefore does not imply classification continuity.

Likewise, classification continuity does not prove identity continuity: two independent Products may both be Primary fraud-detection AI Products without being the same Product identity.

M14-R03 - Identity/classification separation: Continuing identity MUST NOT be inferred from shared classification, label, capability, dependency or implementation similarity.

M14-R04 - Identity-continuity restraint: Continuity of identity MUST NOT force continuity of Product Family, APT, primitive type, capability, risk, lifecycle or other version-specific classifications.


330. AIPS Product Identity and AI Product Version​

AIPS explicitly distinguishes continuing Product identity from immutable Product-version semantics.

The core model is:

Product Identity
|
+---- hasVersion ----> AI Product Version v1
|
+---- hasVersion ----> AI Product Version v2
|
+---- hasVersion ----> AI Product Version v3

The versions do not overwrite one another.

For AIPS-governed AI Products:

  • Product ID provides the continuing logical identity;
  • the exact Product version identifies the version-specific Product state;
  • AIPROD describes one immutable version-specific semantic descriptor;
  • AQE establishes why that exact version qualifies for the AI Product Family;
  • APT and other Product ontology dimensions apply to the exact governed version where specified by AIPS;
  • later evidence/currentness changes do not rewrite the historical version descriptor.

M14 does not broaden aips:hasVersion to all GEAIRM resources.

M14-R05 - Product version immutability: Published AIPS Product-version semantic descriptors and historical qualification/classification evidence MUST NOT be rewritten in place to make later reality appear historically true.

M14-R06 - Product property preservation: GEAIRM MUST NOT broaden AIPS hasVersion into a universal version property for Agents, AI Primitives, Tools, Knowledge Resources or other non-Product participants.


331. Versioned Resource as a Contextual Role​

GEAIRM needs to discuss versioned non-Product resources without pretending that every such resource is an AIProductVersion.

M14 therefore introduces Versioned Resource as a contextual Role.

331.1 Canonical definition​

Versioned Resource is a contextual Role played by an independently identifiable participant when a governed revision, snapshot or release of that participant is represented with sufficient identity and change semantics to distinguish it from other revisions of the continuing resource.

Possible examples include:

  • AI Primitive revision;
  • model artifact release;
  • Knowledge Resource version;
  • Rule set version;
  • Prompt/template version;
  • Tool/service interface version;
  • Agent profile version;
  • Evaluation Report version;
  • Product Version, where Product authority governs the identity.

Playing the Versioned Resource Role does not make the resource a Product or AI Product.

331.2 Element record​

FieldValue
Preferred labelVersioned Resource
Element kindRole
DefinitionA contextual Role played by a resource represented as a governed revision, snapshot or release distinguishable from other revisions of its continuing identity.
Semantic authorityGEAIRM
Representation dispositionOWNED
Normative statuscanonical
Source sliceM14
Constraints / invariantsVersioned Resource is not AIProductVersion, Product status, immutable-by-definition in every external scheme, or a mandatory role for all participants.

M14-R07 - Non-Product versioning boundary: Playing the Versioned Resource Role MUST NOT establish Product, AI Product or AIProductVersion identity.


332. Version Identifier​

M14 introduces Version Identifier as the minimal generic concept required to distinguish governed revisions without mandating SemVer universally.

332.1 Canonical definition​

Version Identifier is a governed identifier or value that distinguishes a represented revision, snapshot or release of a continuing resource within its applicable versioning scheme.

Examples may include:

1.2.0
2026-10-06
revision-17
model-hash-abc123
ontology-0.1

The syntax alone does not determine compatibility or semantic significance.

AIPS recommends semantic versioning for AI Products unless another governed organizational scheme is declared, but GEAIRM does not require SemVer for every enterprise-AI participant.

332.2 identifiedByVersion​

FieldValue
Preferred labelidentified by Version
DefinitionRelates a resource playing Versioned Resource Role to the Version Identifier that distinguishes the represented revision under the declared versioning scheme.
Semantic authorityGEAIRM
Subject kindsresource playing Versioned Resource Role
Object kindsVersion Identifier
Directionalitydirected
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot universally asserted by GEAIRM
Inversenone normatively established by M14
Source sliceM14

M14-R08 - Version-scheme neutrality: GEAIRM MUST NOT require semantic versioning for every enterprise-AI resource merely because AIPS recommends SemVer for Products.

M14-R09 - Version-identifier restraint: Shared or similar version strings MUST NOT be used to infer shared identity, compatibility, succession or semantic equivalence.


333. Versioning Does Not Mean Mutation​

A versioned resource can evolve in two fundamentally different ways:

Historical revision remains represented
+
new revision is created

versus:

same historical record is edited in place

For published governed semantics, M14 strongly prefers the first pattern whenever the change would alter historical meaning, qualification, material dependency, governance posture or provenance.

This does not mean every mutable runtime value requires a new semantic version. Ephemeral runtime state, observations and currentness assessments may change without minting a new Product version.

The controlling distinction is:

semantic identity/history change
-> versioned governed evolution

runtime/currentness observation
-> may update evidence/state without rewriting history

M14-R10 - Historical non-mutation: Material semantic change MUST NOT be represented by silently mutating a historical version whose prior semantics are required for audit, lineage, reproducibility or dependency interpretation.


334. Lifecycle State​

M14 introduces Lifecycle State as a generic M03 State construct for enterprise-AI participants.

334.1 Canonical definition​

Lifecycle State is a context- and time-qualified State describing where a governed participant currently stands in its applicable introduction, operation, deprecation, withdrawal or historical lifecycle.

M14 intentionally does not define one universal Lifecycle State vocabulary.

Different authorities may legitimately define different lifecycle vocabularies for different semantic concerns.

334.2 Element record​

FieldValue
Preferred labelLifecycle State
Element kindState
DefinitionA context- and time-qualified State describing the standing of a governed participant within its applicable lifecycle.
Semantic authorityGEAIRM for generic structure; value vocabulary remains profile/authority-specific
Representation dispositionOWNED
Normative statuscanonical
Source sliceM14
Constraints / invariantsLifecycle State is not intrinsic type, Product Family, Currentness, PDEP stage, Marketplace listing state or assurance status.

334.3 hasLifecycleState​

FieldValue
Preferred labelhas Lifecycle State
DefinitionRelates a governed participant to an applicable Lifecycle State in a declared temporal and authority context.
Semantic authorityGEAIRM
Subject kindsgoverned participant
Object kindsLifecycle State
Directionalitydirected
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot universally asserted across all authorities/contexts
Inversenone normatively established by M14
Source sliceM14

M14-R11 - Lifecycle-state contextuality: Lifecycle State MUST be interpreted relative to the governing lifecycle authority and temporal context and MUST NOT be treated as an intrinsic semantic type.


335. AIPS Lifecycle Stage​

AIPS defines the Product lifecycle vocabulary:

Experimental
Beta
Production
Deprecated
Retired

The current AIPS core defines aips:LifecycleStage and aips:hasLifecycleStage in the Product ontology. M14 reuses these AIPS semantics for AIPS Product lifecycle representation and does not alter their machine-readable domain.

The AIPS stages answer a Product maturity/availability lifecycle question. They are not Product Family, APT, risk tier or currentness status.

For example:

Production + qualification STALE

is semantically possible during a currentness incident until governance acts.

Likewise:

Retired + historical qualification CURRENT-at-evaluation-time

does not mean the Product version is currently available for consumption.

M14-R12 - AIPS lifecycle authority preservation: GEAIRM MUST preserve AIPS authority for Lifecycle Stage and its current controlled values.

M14-R13 - Lifecycle/qualification distinction: AIPS Lifecycle Stage MUST remain distinct from AI Product Family qualification and qualification currentness.


336. Lifecycle State, Process Stage, Listing State and Assurance Status Are Different​

The AIPS ecosystem contains several state-like vocabularies that answer different questions.

Semantic concernExample vocabularyQuestion answered
AIPS Lifecycle StageExperimental, Beta, Production, Deprecated, RetiredWhere is the Product in its lifecycle?
PDEP Process Stagenumbered lifecycle-development stagesWhich governed development/evolution step is being executed?
Marketplace Listing StateActive, Restricted, Suspended, Historical, RetiredHow is this exact version currently represented for discovery/consumption?
Qualification StatusCURRENT, STALE, REVALIDATION_REQUIRED, INVALIDIs AI Product qualification evidence currently valid?
PDEP Continuous Assurance statusHEALTHY, ATTENTION_REQUIRED, REVALIDATION_REQUIRED, SUSPENDEDWhat is the current continuous-assurance operational status?
COVP currentnessCOVP-specific currentness valuesIs consumption evidence/current assessment still suitable in its own scope?

These vocabularies may influence one another through governed rules, but they are not interchangeable.

Retired appearing in more than one vocabulary does not establish semantic identity merely from the shared label.

M14-R14 - State-vocabulary separation: Lifecycle Stage, PDEP process stage, Marketplace listing state, Qualification Status, Continuous Assurance status and COVP currentness MUST remain distinct semantic dimensions.

M14-R15 - Label collision restraint: Identical or similar labels across lifecycle/currentness vocabularies MUST NOT be treated as the same semantic identity without explicit authority mapping.


337. Currentness Assessment​

Currentness is not an intrinsic identity property. It is a time- and concern-qualified judgment about whether a previously established semantic/evidential state remains suitable for a declared present purpose.

M14 introduces Currentness Assessment.

337.1 Canonical definition​

Currentness Assessment is a time- and concern-qualified assessment of whether specified semantics, evidence, dependencies, governance or assurance for a declared target remain sufficiently up to date and applicable for the stated current purpose.

Examples of currentness concerns include:

  • AI qualification;
  • dependencies;
  • governance profile;
  • assurance evidence;
  • fitness evidence;
  • marketplace trust publication;
  • lineage/dependency binding;
  • policy applicability.

337.2 Element record​

FieldValue
Preferred labelCurrentness Assessment
Element kindConcept / assessment state record
DefinitionA time- and concern-qualified assessment of whether declared semantics/evidence remain sufficiently current and applicable for a stated purpose.
Semantic authorityGEAIRM generic structure; status vocabulary may be externally governed
Representation dispositionOWNED
Normative statuscanonical
Source sliceM14
Constraints / invariantsCurrentness Assessment is not intrinsic identity, Lifecycle State, Product version, historical validity or universal freshness timestamp.

337.3 Relationships​

M14 introduces:

assessesCurrentnessOf
Currentness Assessment -> target

hasCurrentnessConcern
Currentness Assessment -> concern/category

hasCurrentnessStatus
Currentness Assessment -> status value under applicable authority

The generic GEAIRM structure does not impose one universal currentness status vocabulary.

M14-R16 - Currentness contextuality: Currentness MUST be interpreted relative to a target, concern, evaluation time and applicable authority rather than as an intrinsic timeless property.

M14-R17 - Currentness-vocabulary restraint: GEAIRM MUST NOT generalize one profile's currentness status vocabulary into a universal vocabulary for all currentness concerns.


338. AIPS Qualification Currentness​

AIPS uses QualificationStatus to distinguish historical qualification evidence from currently usable qualification evidence.

The current conceptual values are:

CURRENT
STALE
REVALIDATION_REQUIRED
INVALID

Their role is specific:

  • CURRENT — evidence remains valid for the declared Product version and bound dependencies;
  • STALE — relevant upstream or constitutional information changed and evidence may no longer reflect current reality;
  • REVALIDATION_REQUIRED — material change requires explicit recomputation or governance review;
  • INVALID — the Product version no longer satisfies AI Product qualification.

Historical evidence is retained even when currentness changes.

This yields an important distinction:

historically qualified
!= currently qualified for present discovery/consumption

An INVALID currentness status does not mean historical records should be erased. It means the version must not be represented as a current qualifying AI Product under AIPS rules.

M14-R18 - Qualification-currentness authority: GEAIRM MUST preserve AIPS QualificationStatus semantics and MUST NOT redefine CURRENT/STALE/REVALIDATION_REQUIRED/INVALID for unrelated currentness domains.

M14-R19 - Historical qualification preservation: Loss of current qualification MUST NOT erase the historical fact that a prior evaluation established qualification for the historical version/context.


339. PDEP Trust Currentness​

PDEP defines TrustCurrentness as a version-specific evidence object with separate dimensions:

qualificationCurrentness
dependencyCurrentness
governanceCurrentness
assuranceCurrentness

The current PDEP 0.1 machine model uses the AIPS QualificationStatus range for all four dimensions.

M14 does not generalize this implementation choice beyond the PDEP profile.

The semantic distinction remains:

qualification currentness
Is Product Family qualification evidence still current?

dependency currentness
Are material/version bindings and relevant upstream assumptions current?

governance currentness
Is the applicable governance compilation/profile current?

assurance currentness
Is the assurance evidence/finding current enough for the declared trust use?

One dimension may degrade while another remains current.

For example:

qualification = CURRENT
dependencies = CURRENT
governance = REVALIDATION_REQUIRED
assurance = CURRENT

is not contradictory.

M14-R20 - Trust-currentness dimensionality: Qualification, dependency, governance and assurance currentness MUST remain independently representable and MUST NOT be collapsed into one scalar unless an authoritative profile defines an aggregate status.


340. Dependency Currentness​

M11 established that dependency semantics are version-sensitive and that upstream publication does not itself rewrite downstream bindings.

M14 adds the currentness distinction:

upstream has a newer version
!= downstream dependency is automatically stale

A downstream binding may remain current when:

  • it intentionally pins the older version;
  • that version remains permitted and available;
  • no policy or compatibility rule requires migration;
  • the governed assumptions supporting the binding remain valid.

Conversely, a dependency may become stale or require revalidation even without a new upstream version, for example when:

  • the upstream version loses qualification currentness;
  • governance policy changes;
  • compatibility evidence is invalidated;
  • an entitlement or safety assumption changes;
  • an upstream Product is suspended or retired for current use.

AIPS DependencyCurrentness is also a Governance Obligation. The obligation to monitor dependency currentness is distinct from the currentness assessment itself.

M14-R21 - New-version/currentness distinction: Publication of a newer upstream version MUST NOT by itself make a pinned downstream dependency stale.

M14-R22 - Obligation/currentness distinction: AIPS DependencyCurrentness Governance Obligation MUST remain distinct from the actual assessment/state of dependency currentness.


341. Governance and Assurance Currentness​

Governance currentness asks whether the currently applicable governance compilation, obligations, policies and supporting evidence still correspond to the Product/version and its context.

Assurance currentness asks whether assurance evidence and findings remain sufficiently applicable to the current Product/version, dependencies, intended use and operating conditions.

Neither is equivalent to historical existence of a Governance Obligation Profile or Assurance record.

For example:

Governance Obligation Profile exists
!= governance currentness CURRENT

Assurance PASS recorded six months ago
!= assurance currentness CURRENT today automatically

Currentness can degrade because the target, dependency, policy, environment, intended use or evidence basis changed.

M14-R23 - Record/currentness distinction: Existence of governance or assurance records MUST NOT be treated as evidence that those records are currently applicable.


342. Currentness Does Not Rewrite Historical Truth​

M14 establishes the core invariant:

Historical semantic truth is preserved; currentness is reassessed.

Suppose:

Product A v1
qualified as AI Product
APT = Primary
depended on Model Product M1

Later:

Product A v2
replaces M1 with ordinary deterministic logic
no qualifying AI Primitive remains

Then:

A v1 remains historically an AI Product
A v2 is assessed from its own constitution

The later Product reality does not rewrite v1.

Likewise, a currentness assessment may change from CURRENT to STALE without changing what the historical version was. It changes whether the evidence can still be relied upon for a current decision.

M14-R24 - Historical/current distinction: Currentness change MUST NOT be interpreted as retroactive mutation of the historical semantic state of the governed version.


343. Supersession​

M14 introduces Supersession as context-bounded relationship semantics needed to distinguish replacement for current use from historical deletion.

343.1 Canonical definition​

Supersession is a governed relationship in which a later resource, version, evidence item or projection replaces an earlier one for a declared current purpose while the superseded resource remains historically identifiable.

Supersession may apply to:

  • Product versions;
  • Qualification Evidence;
  • semantic descriptors;
  • marketplace projections;
  • Evaluation Reports;
  • assurance findings;
  • other governed versioned resources.

Supersession does not imply that the later resource is semantically equivalent to the earlier resource.

343.2 supersedes​

FieldValue
Preferred labelsupersedes
DefinitionRelates a later governed resource to an earlier resource that it replaces for a declared current purpose while preserving historical identity of the earlier resource.
Semantic authorityGEAIRM generic relationship; narrower AIPS fields/properties remain authoritative in their own scopes
Subject kindsgoverned resource/version/evidence/projection
Object kindsgoverned resource/version/evidence/projection
Directionalitydirected: later -> earlier
Transitivitynot universally asserted
Symmetryfalse
Functional cardinalitynot asserted
Inversenot normatively minted in M14
Source sliceM14

AIPS already defines narrower supersession semantics such as supersedesQualificationEvidence, and Marketplace records lifecycle links such as supersedes / supersededBy. Those scoped mechanisms remain authoritative.

M14-R25 - Supersession/history preservation: Supersession MUST NOT erase, mutate or semantically overwrite the superseded resource's historical record.

M14-R26 - Supersession/equivalence distinction: Supersession MUST NOT imply semantic equivalence, backward compatibility, same Product Family or same qualification status.


344. Deprecation, Retirement and Historical Closure​

Deprecation and retirement concern lifecycle availability/standing. They do not negate historical identity or provenance.

AIPS lifecycle documentation requires retired Product versions to retain AIPROD, qualification evidence and lineage for audit.

AIPS Lifecycle Closure further defines Historical Closure as the state in which a superseded or retired Product version retains immutable historical ontology and trust evidence while no longer being represented as the current consumable version.

The distinctions are:

Deprecated
still represented but scheduled for retirement

Retired
no longer available for new consumption under Product lifecycle semantics

Historical listing/state
discoverability projection for historical evidence

Historical Closure
cross-artifact evidence that historical semantics remain coherent

These concepts may correlate but are not interchangeable.

M14-R27 - Retirement/history distinction: Retirement MUST NOT imply deletion of historical identity, qualification evidence, lineage, provenance or audit records required by the governing policy.

M14-R28 - Historical-closure distinction: Historical Closure MUST remain distinct from Product lifecycle Retired state and Marketplace Historical listing state.


345. Provenance​

M14 does not mint a GEAIRM Provenance superclass.

W3C PROV remains the generic external semantic foundation for provenance entities, activities, derivation and attribution, while AIPS specializes provenance for Product, qualification, risk, governance and lifecycle evidence.

M14 uses provenance as the semantic concern answering:

What origins, derivations, activities, agents, evidence and transformations explain how this governed resource or semantic state came to be?

For AIPS Product semantics, provenance may include:

  • AI Primitives and Model Products;
  • Data Products and sources;
  • Knowledge, rules, prompts and policies;
  • Tools and services;
  • upstream AI Products;
  • contributors and accountable organizations;
  • creation/release events;
  • qualification evidence;
  • superseded/derived versions.

M14 preserves AIPS derivedFrom as a provenance relationship and does not treat it as synonymous with material dependency.

M14-R29 - Provenance non-duplication: GEAIRM MUST NOT create a shadow provenance hierarchy where PROV/AIPS already provide authoritative provenance semantics.

M14-R30 - Derivation/dependency distinction: Provenance derivation such as derivedFrom MUST remain distinct from dependsOn and materiallyDependsOn; derivation history does not automatically establish current operational/material dependency.


346. Lineage​

AIPS defines Lineage as the record of dependencies, upstream/downstream Products and version history.

M14 treats Lineage as a graph/view over explicit identity, version, dependency and provenance semantics rather than minting a second generic lineage entity class.

Lineage answers questions such as:

  • which exact upstream versions contributed to this version?
  • which dependency path supported qualification?
  • what version history led here?
  • which downstream versions are actually bound to this version?
  • which semantic changes occurred across versions?
  • what provenance explains those changes?

The distinction is:

Dependency
what a participant relies on

Provenance
how a participant/version came to be

Lineage
traceable graph/view across origin, dependency and version history

A lineage edge may represent dependency, derivation, succession or another explicit relationship. The word lineage alone does not authorize inference of one relationship type from another.

M14-R31 - Lineage relationship discipline: A lineage view MUST preserve the semantics of its underlying relationship types and MUST NOT collapse dependency, derivation, supersession and version succession into one generic edge.


347. Version Binding and Historical Dependency Truth​

M11 established that dependency bindings should be sufficiently precise where qualification/governance semantics depend on them.

M14 strengthens the historical invariant:

Downstream D v7 -> Upstream U v3

means that the historical semantics of D v7 must be interpreted against U v3 unless another authoritative record states otherwise.

Publishing U v4 does not change the meaning of D v7.

If a new downstream version D v8 adopts U v4, D v8 receives its own qualification, APT, governance and assurance assessment as required.

This enables reproducibility and impact analysis without cascading historical mutation.

M14-R32 - Bound-version interpretation: Historical downstream semantics MUST be interpreted against the dependency versions actually bound to that downstream version, not automatically against the latest available upstream version.


348. Semantic Change Event​

M14 introduces Semantic Change Event as a minimal generic concept for a material change that may require currentness reassessment, new versioning or downstream impact analysis.

348.1 Canonical definition​

Semantic Change Event is a recorded change occurrence that may alter the governed meaning, constitution, qualification, dependency assumptions, behaviour, effect, authority, risk, governance, assurance or lifecycle interpretation of a participant.

Examples include:

  • AI Primitive addition/removal/replacement;
  • upstream AI Product version adoption;
  • Product Family reclassification;
  • APT change;
  • material capability/effect/autonomy/embodiment/topology change;
  • governance policy change;
  • qualification invalidation;
  • dependency-currentness degradation;
  • retirement or supersession.

A Semantic Change Event does not by itself say whether the change is breaking, major/minor/patch, or requires a new Product identity.

348.2 affectsSemanticTarget​

FieldValue
Preferred labelaffects semantic Target
DefinitionRelates a Semantic Change Event to a governed participant/version whose semantic interpretation or currentness may require reassessment because of that change.
Semantic authorityGEAIRM
Subject kindsSemantic Change Event
Object kindsgoverned participant/version
Directionalitydirected
Transitivitynot asserted
Symmetryfalse / not asserted
Functional cardinalitynot asserted
Inversenone normatively established by M14
Source sliceM14

M14-R33 - Change-event restraint: A Semantic Change Event MUST NOT by itself determine version increment level, Product identity break, reclassification outcome or lifecycle action.


349. Change Impact and Controlled Reassessment​

AIPS treats ontology-affecting change as a versioned graph event, not an in-place label edit.

M14 adopts the generic pattern:

Semantic Change Event
|
v
identify actual bound/material paths
|
v
reassess affected currentness / semantics
|
+---- no material impact ---> preserve current bindings/state
|
+---- material impact ------> governed re-entry / new evidence /
new version where required

Impact propagates through actual adopted/material bindings rather than names, ancestry or mere existence of a newer version.

This means:

same Product name
!= affected automatically

same upstream identity
!= adopted newest version automatically

transitive ancestry
!= current material impact automatically

M14-R34 - Change-impact binding discipline: Downstream impact MUST be determined from actual version bindings, material dependency paths and applicable governance rules rather than from names or ancestry alone.


350. Reclassification Across Versions​

A continuing Product identity may survive a version whose Product Family or APT changes, where BPS/AIPS identity rules permit that continuity.

AIPS explicitly allows the example:

Fraud Assessment Product
v1: AI Product using learned-model primitive
v2: conventional rules/software Product

The correct interpretation is:

same continuing Product identity may remain
historical v1 classification remains intact
v2 receives its own intrinsic classification

Likewise APT may evolve:

Primary -> Derived
Derived -> Primary

across versions when the intelligence constitution changes.

Reclassification does not mean the old classification was an error. It means the later version has a different constitution/current semantic state.

M14-R35 - Version-specific reclassification: Product Family and APT reclassification MUST be applied to the applicable version and MUST NOT retroactively relabel historical versions.


351. Currentness and Discovery / Consumption​

AIPS Marketplace requires discovery to reflect the current truth of a specific Product version.

The current conceptual listing states are:

Active
Restricted
Suspended
Historical
Retired

AIPS explicitly prohibits Active/Restricted current-AI-Product representation when qualification is STALE, REVALIDATION_REQUIRED or INVALID.

M14 preserves this as Marketplace authority rather than redefining listing policy generically.

The semantic pattern is:

historical semantic identity
remains available

current trust/currentness state
controls present eligibility

Entitlement also does not override currentness: a previously granted entitlement does not force continued access when the Product is suspended, retired, revoked or no longer satisfies current policy.

M14-R36 - Discovery/currentness separation: Marketplace listing state MUST remain a discovery/consumption projection and MUST NOT redefine Product Family, APT or historical identity.

M14-R37 - Entitlement/currentness distinction: Entitlement MUST NOT be interpreted as proof that a Product version remains current, trusted or eligible for present consumption.


352. Currentness Is Not Freshness Alone​

A timestamp can help establish recency, but currentness is semantic rather than merely chronological.

A year-old evidence record may remain current if:

  • the target is unchanged;
  • dependencies remain valid;
  • policy remains applicable;
  • assurance assumptions remain satisfied.

A five-minute-old record may already be stale if a material dependency or policy changed immediately afterward.

Therefore:

newer timestamp != more current automatically
older timestamp != stale automatically

Currentness depends on the relation between evidence and present governed reality.

M14-R38 - Currentness/freshness distinction: Currentness MUST NOT be reduced to age or timestamp recency alone.


353. Currentness Is Not Validity, Availability or Fitness​

Several commonly conflated notions remain distinct:

Currentness
evidence/semantics remain applicable now

Validity
artifact/assertion satisfies applicable validity rules

Availability
resource can currently be accessed/used

Fitness for Purpose
Product version is suitable for a particular intended use

Lifecycle State
participant's lifecycle standing

A Product can be:

  • current but unavailable because access is restricted;
  • valid but stale for a changed intended use;
  • historically valid but no longer current;
  • current but UNFIT for a particular intended use;
  • retired yet historically coherent.

M14-R39 - Currentness-axis separation: Currentness MUST remain distinct from validity, availability, fitness for purpose and Lifecycle State.


354. Provenance and Evidence Are Separate​

M13 established Evidence as a contextual Role.

Provenance explains origin and derivation. Evidence is a role played by a resource when used to support or challenge a claim/assessment.

A provenance record may play Evidence Role in an Evaluation or Assurance Assessment, but:

Provenance != Evidence Role

Likewise, an Evaluation Report may be Evidence without itself providing the complete provenance of the evaluated Product.

M14-R40 - Provenance/evidence distinction: Provenance semantics and M13 Evidence Role MUST remain distinct even when provenance records are used as evidence.


355. Identity, Version and Lifecycle for Non-Product AI​

GEAIRM applies to enterprise AI more broadly than Productized AI.

An Agent, AI Primitive, Knowledge Resource, Tool, Composition or Evaluation may have identity, version, lifecycle and provenance semantics without becoming a Product.

For example:

PolicyReasoner
identity: urn:reasoner:policy
version: 3.2
lifecycle state: governed under its own profile
provenance: derived from ruleset 18 + engine release 7

Nothing in those assertions establishes:

  • BPS Product qualification;
  • AI Product Family membership;
  • APT;
  • AIPS Product Lifecycle Stage;
  • R0-R4 risk tier.

Those Product semantics require the applicable Product authority.

M14-R41 - Non-Product lifecycle boundary: Identity, version, lifecycle or provenance semantics applied to a non-Product participant MUST NOT by themselves establish Product or AI Product status.


356. Canonical M14 Reference View​

                    CONTINUING IDENTITY

+--------------------------+
| governed participant |
| continuing identity |
+------------+-------------+
|
represented through
governed revisions
|
+----------------+----------------+
| |
v v
Versioned Resource v1 Versioned Resource v2
identifiedByVersion identifiedByVersion
| |
v v
Version ID 1 Version ID 2

v2 -- supersedes --> v1 [where explicitly governed]

Historical v1 remains identifiable and semantically preserved.


AIPS PRODUCT SPECIALIZATION

Product Identity
|
+---- hasVersion ----> AI Product Version v1
|
+---- hasVersion ----> AI Product Version v2

Each version has its own AQE / APT / AIPROD / governance / lineage.


LIFECYCLE VS CURRENTNESS

Participant/version
|
+---- hasLifecycleState ----> Lifecycle State
|
+---- hasCurrentnessAssessment ---> Currentness Assessment
|
+-----------+------------+
| | |
target concern status

AIPS / PDEP may specialize currentness into:
qualification | dependencies | governance | assurance

Lifecycle Stage != Currentness Status != Marketplace Listing State


PROVENANCE & LINEAGE

PROV / AIPS provenance
explains origin / derivation / generation

Dependency graph
explains current or historical reliance

Version history
explains revision/succession

Lineage view
traces these explicit relationships
without collapsing their meanings.


CHANGE

Semantic Change Event
|
v
actual bound/material dependency paths
|
v
Currentness / semantic reassessment
|
+---- preserve existing version where unaffected
|
+---- new evidence / lifecycle action / new version where required

The view is semantic. It does not prescribe registry, source-control, event-store, artifact-repository or lifecycle-engine architecture.


357. Illustrative Semantic Examples​

357.1 AI Product changes Product Family across versions​

Product Identity Fraud-Assessment

v1
Product Family = AI Product
APT = Primary
qualifying primitive = Learned Model M1

v2
constitution = ordinary deterministic rules/software
no qualifying AI Primitive path

v1 remains historically an AI Product. v2 is classified from its own PIN and constitution.

357.2 New upstream version does not mutate downstream history​

Upstream A v1
Downstream X v7 -> A v1

later:
Upstream A v2 published

X v7 remains interpreted against A v1 unless X v7's governed record actually changes, which would itself violate the historical immutability principle for a published version. A new X v8 may adopt A v2.

357.3 Qualification becomes stale without rewriting Product version​

Product Version P4
immutable AIPROD
AQE qualificationStatus = CURRENT

material upstream evidence changes

new Currentness Assessment
qualification = STALE

The historical AIPROD remains intact while current discovery/consumption policy may change.

357.4 Product lifecycle and marketplace state differ​

AIPS Lifecycle Stage = Production
Marketplace Listing State = Suspended

This can occur during a trust/currentness incident. Production maturity does not force Active listing.

357.5 Retired version remains historically discoverable​

Product Version P2
Lifecycle = Retired
Marketplace = Historical
Lifecycle Consistency = HISTORICAL_CLOSED

These are related but distinct state assertions.

357.6 Non-Product primitive versioning​

AI Primitive Reasoner-X
revision 4.3
provenance: derived from rule-engine release 9

Versioning/provenance does not make Reasoner-X a Product.

357.7 Provenance is not dependency​

Evaluation Report ER2
derivedFrom ER1

This provenance relationship does not mean ER2 operationally dependsOn ER1 at runtime.

357.8 Old assurance evidence becomes stale​

Assurance Finding AF1
produced at T1
status then applicable

Product environment changes materially at T2

Currentness Assessment at T3
assurance concern = REVALIDATION_REQUIRED

AF1 remains historical evidence; its present applicability changed.


358. M14 Canonical Element Inventory​

ElementM03 kindAuthorityDispositionM14 status
Product IdentityConceptAIPS/BPSREUSEDcanonical external
AI Product VersionConcept / versioned Product entityAIPSREUSEDcanonical external
hasVersionRelationship TypeAIPSREUSEDcanonical external
Versioned ResourceRoleGEAIRMOWNEDcanonical
Version IdentifierConcept / identifying valueGEAIRMOWNEDcanonical
identifiedByVersionRelationship TypeGEAIRMOWNEDcanonical
Lifecycle StateStateGEAIRMOWNEDcanonical
hasLifecycleStateRelationship TypeGEAIRMOWNEDcanonical
AIPS Lifecycle StageClassificationAIPSREUSEDcanonical external
Experimental / Beta / Production / Deprecated / RetiredClassification valuesAIPSREUSEDcanonical external
PDEP Process Stageprocess-state conceptAIPS PDEPREFERENCEDcanonical external
Currentness AssessmentConcept / assessment-state recordGEAIRMOWNEDcanonical
assessesCurrentnessOfRelationship TypeGEAIRMOWNEDcanonical
hasCurrentnessConcernRelationship TypeGEAIRMOWNEDcanonical
hasCurrentnessStatusRelationship TypeGEAIRMOWNEDcanonical
Qualification StatusClassificationAIPSREUSEDcanonical external
CURRENT / STALE / REVALIDATION_REQUIRED / INVALIDClassification valuesAIPSREUSED in AIPS currentness scopescanonical external
Trust Currentnessevidence/state recordAIPS PDEPREUSED / REFERENCEDcanonical external
Marketplace Listing Statediscovery stateAIPS MarketplaceREFERENCEDcanonical external
Continuous Assurance statusoperational assurance statusAIPS PDEPREFERENCEDcanonical external
COVP currentnessconsumption-evidence currentnessAIPS COVPREFERENCEDcanonical external
SupersessionRelationship semanticsGEAIRMOWNEDcanonical
supersedesRelationship TypeGEAIRMOWNEDcanonical generic relation
Historical Closurelifecycle consistency stateAIPSREUSED / REFERENCEDcanonical external
Provenanceexternal semantic domainW3C PROV + AIPS specializationREUSED / REFERENCEDno GEAIRM shadow class
derivedFromRelationship TypeAIPS / PROVREUSEDcanonical external
Lineagegraph/view termAIPS / GEAIRM editorial integrationREFERENCED / not new classcanonical semantic view
Semantic Change EventConceptGEAIRMOWNEDcanonical
affectsSemanticTargetRelationship TypeGEAIRMOWNEDcanonical

No executable GEAIRM identifiers are minted in M14. Stable machine identifiers remain deferred to the A1-A8 formalization programme after M17.


359. M14 Hard Distinctions​

The following distinctions are normative for subsequent slices:

identity != label
identity != classification
identity continuity != classification continuity
classification similarity != identity continuity

Product Identity != AI Product Version
AI Product Version != generic Versioned Resource Role
Version Identifier != identity
Version Identifier != compatibility guarantee

Lifecycle State != intrinsic type
Lifecycle State != Currentness
AIPS Lifecycle Stage != PDEP process stage
AIPS Lifecycle Stage != Marketplace listing state
Marketplace listing state != Qualification Status
Continuous Assurance status != Lifecycle Stage

Currentness != historical validity
Currentness != freshness alone
Currentness != availability
Currentness != Fitness for Purpose
Currentness != Lifecycle State

Qualification currentness != dependency currentness
qualification currentness != governance currentness
qualification currentness != assurance currentness
DependencyCurrentness obligation != dependency currentness assessment

supersession != deletion
supersession != semantic equivalence
supersession != historical mutation
Retired != Historical listing state
Retired != Historical Closure

Provenance != Dependency
Provenance != Evidence Role
Lineage != one generic edge type
derivedFrom != materiallyDependsOn

new upstream version != automatic downstream adoption
new upstream version != automatic downstream staleness
currentness change != historical ontology mutation
reclassification != admission that historical classification was wrong

non-Product versioning != Product qualification

As in prior slices, != denotes semantic non-equivalence rather than formal OWL disjointness unless later executable formalization explicitly authorizes stronger axioms.


360. M14 Normative Rules​

RuleRequirement
M14-R01GEAIRM MUST preserve AIPS authority for ProductIdentity, AIProductVersion and hasVersion in Product semantics.
M14-R02GEAIRM MUST reuse/reference W3C PROV and AIPS provenance semantics rather than create a competing generic provenance ontology.
M14-R03Continuing identity MUST NOT be inferred solely from shared classification, label, capability, dependency or implementation similarity.
M14-R04Identity continuity MUST NOT force continuity of Product Family, APT, primitive type, capability, risk, lifecycle or other version-specific classifications.
M14-R05Published AIPS Product-version semantic descriptors and historical qualification/classification evidence MUST NOT be rewritten in place to reflect later reality.
M14-R06GEAIRM MUST NOT broaden AIPS hasVersion into a universal version property for non-Product resources.
M14-R07Playing Versioned Resource Role MUST NOT establish Product, AI Product or AIProductVersion identity.
M14-R08GEAIRM MUST NOT require semantic versioning for every enterprise-AI resource merely because AIPS recommends it for Products.
M14-R09Shared/similar version identifiers MUST NOT imply shared identity, compatibility, succession or semantic equivalence.
M14-R10Material semantic change MUST NOT silently mutate a historical governed version where prior semantics are required for audit, lineage, reproducibility or dependency interpretation.
M14-R11Lifecycle State MUST be interpreted relative to governing lifecycle authority and temporal context and MUST NOT become an intrinsic type.
M14-R12GEAIRM MUST preserve AIPS authority for Lifecycle Stage and its current controlled values.
M14-R13AIPS Lifecycle Stage MUST remain distinct from AI Product Family qualification and qualification currentness.
M14-R14Lifecycle Stage, PDEP process stage, Marketplace listing state, Qualification Status, Continuous Assurance status and COVP currentness MUST remain distinct.
M14-R15Shared state labels across different vocabularies MUST NOT establish semantic identity without explicit authority mapping.
M14-R16Currentness MUST be interpreted relative to target, concern, evaluation time and applicable authority rather than as a timeless intrinsic property.
M14-R17GEAIRM MUST NOT generalize one profile's currentness status vocabulary into a universal currentness vocabulary.
M14-R18GEAIRM MUST preserve AIPS QualificationStatus semantics and MUST NOT redefine CURRENT/STALE/REVALIDATION_REQUIRED/INVALID for unrelated domains.
M14-R19Loss of current qualification MUST NOT erase the historical fact of earlier qualification for the historical version/context.
M14-R20Qualification, dependency, governance and assurance currentness MUST remain independently representable unless an authoritative profile defines an aggregate status.
M14-R21Publication of a newer upstream version MUST NOT by itself make a pinned downstream dependency stale.
M14-R22AIPS DependencyCurrentness Governance Obligation MUST remain distinct from dependency currentness assessment/state.
M14-R23Existence of governance or assurance records MUST NOT prove that those records remain currently applicable.
M14-R24Currentness change MUST NOT be interpreted as retroactive mutation of the historical semantic state of a governed version.
M14-R25Supersession MUST preserve the historical record of the superseded resource.
M14-R26Supersession MUST NOT imply semantic equivalence, backward compatibility, same Product Family or same qualification status.
M14-R27Retirement MUST NOT imply deletion of historical identity, qualification evidence, lineage, provenance or required audit records.
M14-R28Historical Closure MUST remain distinct from Product Retired state and Marketplace Historical listing state.
M14-R29GEAIRM MUST NOT create a shadow provenance hierarchy where PROV/AIPS already provide authoritative provenance semantics.
M14-R30Provenance derivation MUST remain distinct from dependsOn/materiallyDependsOn and MUST NOT automatically establish current operational/material dependency.
M14-R31A lineage view MUST preserve the semantics of underlying dependency, derivation, supersession and version relationships.
M14-R32Historical downstream semantics MUST be interpreted against the dependency versions actually bound to that downstream version rather than latest available upstream versions.
M14-R33Semantic Change Event MUST NOT by itself determine version increment, Product identity break, reclassification outcome or lifecycle action.
M14-R34Downstream change impact MUST be determined from actual version bindings, material paths and governance rules rather than names or ancestry alone.
M14-R35Product Family/APT reclassification MUST apply to the applicable version and MUST NOT retroactively relabel historical versions.
M14-R36Marketplace listing state MUST remain a discovery/consumption projection and MUST NOT redefine Product Family, APT or historical identity.
M14-R37Entitlement MUST NOT prove that a Product version remains current, trusted or eligible for present consumption.
M14-R38Currentness MUST NOT be reduced to age or timestamp recency alone.
M14-R39Currentness MUST remain distinct from validity, availability, Fitness for Purpose and Lifecycle State.
M14-R40Provenance semantics and M13 Evidence Role MUST remain distinct even when provenance records are used as evidence.
M14-R41Identity, version, lifecycle or provenance semantics applied to a non-Product participant MUST NOT by themselves establish Product or AI Product status.
M14-R42M14 relationships MUST follow M03 conservative inference defaults unless stronger semantics are explicitly established.
M14-R43M14 MUST NOT prescribe registry, source-control, lineage-engine, event-store, artifact-repository or lifecycle-orchestration architecture.
M14-R44M14 semantic non-equivalence MUST NOT be strengthened into formal OWL disjointness unless later executable formalization explicitly authorizes it.

361. M14 Exit Criteria​

GEAIRM-M14 is complete when:

  • continuing identity is explicitly separated from version-specific classification;
  • AIPS ProductIdentity / AIProductVersion / hasVersion authority is preserved;
  • historical Product-version semantic descriptors remain immutable;
  • generic non-Product versioning can be represented without turning resources into Product versions;
  • Version Identifier is defined without imposing SemVer universally;
  • generic Lifecycle State is defined as a context/time-qualified State;
  • AIPS Lifecycle Stage and its current values remain AIPS-owned;
  • Lifecycle Stage is separated from PDEP process stage, Marketplace listing state, Qualification Status, Continuous Assurance status and COVP currentness;
  • generic Currentness Assessment is defined as target/concern/time/authority-qualified;
  • no universal currentness vocabulary is imposed;
  • AIPS QualificationStatus is preserved in its authoritative scope;
  • qualification, dependency, governance and assurance currentness remain independent dimensions;
  • newer upstream publication does not automatically make pinned downstream versions stale;
  • DependencyCurrentness obligation is separated from actual currentness state;
  • currentness change does not rewrite historical truth;
  • Supersession is represented without deletion or equivalence inference;
  • retirement and Historical Closure preserve audit/history semantics;
  • W3C PROV remains the generic provenance foundation;
  • provenance derivation remains distinct from dependency;
  • lineage remains a traceable graph/view over explicit relationship types rather than a generic edge;
  • historical downstream interpretation follows actual version bindings;
  • Semantic Change Event is represented without pre-deciding version increment or reclassification;
  • change impact follows actual material/version bindings;
  • Product Family/APT reclassification is version-specific rather than retroactive;
  • Marketplace currentness affects discovery/consumption without redefining Product ontology;
  • currentness remains distinct from timestamp freshness, availability, validity and Fitness for Purpose;
  • provenance remains distinct from Evidence Role;
  • non-Product identity/version/lifecycle/provenance semantics do not imply Product status;
  • no registry/version-control/lineage/lifecycle architecture is prescribed;
  • M01-M13 authority, abstraction and conservative-inference rules remain intact.

M14 conclusion: these criteria are satisfied by this slice. GEAIRM now has a stable temporal and historical semantic model in which continuing identity, immutable governed versions, lifecycle standing, currentness assessments, supersession, provenance, lineage and change impact remain distinct while integrating cleanly with AIPS Product lifecycle and trust-currentness semantics.


362. Forward Work​

The next slice is:

GEAIRM-M15 - Canonical Relationship Model & Hard Distinctions

M15 will consolidate the relationship semantics introduced across M04-M14, normalize subject/object expectations and inference behaviour, and produce the cross-slice invariant catalogue needed before canonical examples, reference views and executable formalization. It will make explicit which relationships are identity-preserving, context-qualified, occurrence-level, Product-scoped, externally governed or intentionally non-transitive.


363. GEAIRM-M15 - Canonical Relationship Model & Hard Distinctions​

363.1 Purpose of this slice​

M04-M14 established the substantive enterprise-AI concepts and relationships of GEAIRM 0.1. M15 now consolidates those relationships into one canonical relationship model and turns the cross-slice non-equivalence statements into a coherent invariant catalogue.

M15 answers eight questions:

  1. What are the canonical GEAIRM-owned relationship types established so far?
  2. Which externally governed relationships are reused by GEAIRM but remain owned by AIPS, the Agent Ontology or W3C PROV?
  3. What subject/object expectations apply to each relationship?
  4. Which relationships are occurrence-level, context-qualified, Product-scoped, role-based, temporal or externally governed?
  5. What inference behaviour is permitted by default?
  6. Which cross-relationship entailments are explicitly authorized, and which tempting inferences are prohibited?
  7. Which hard distinctions must remain invariant across all later examples, profiles and executable formalization?
  8. Which invariants are strong enough to become candidates for A3 executable/assessment constraints after M17?

M15 is a consolidation slice. It introduces no new enterprise-AI domain concept, Product classification, Agent type, risk taxonomy, lifecycle vocabulary or architecture construct.


364. Consolidation Principles​

The M15 canonical relationship model is governed by seven principles.

364.1 Authority before convenience​

A relationship already owned by AIPS, the KivaNura Agent Ontology, W3C PROV, BPS or another authoritative specification MUST remain externally governed even where GEAIRM uses it in an integrated view.

364.2 Relationship meaning is not diagram meaning​

Spatial proximity, arrow direction in an illustrative diagram, container placement or visual grouping MUST NOT create semantics not explicitly stated by the relationship model.

364.3 Participation does not transfer identity​

A participant can enter an Execution, Context, Composition, Federation, Orchestration, Evaluation, Control Loop or Decision Formation process without inheriting the identity of the process or other participants.

364.4 Occurrence semantics do not become Product declarations​

A property observed or asserted for one AI Execution, action, Evaluation or other occurrence MUST NOT automatically become a declaration about every execution or the governing AI Product Version.

364.5 Product-scoped properties stay Product-scoped​

AIPS properties whose domain is AIProductVersion or another AIPS Product entity MUST NOT be widened merely because a generic GEAIRM concept uses the same controlled vocabulary.

364.6 Context qualifications are semantically material​

Where a relationship is defined as context-bounded, the assertion MUST NOT be generalized beyond the declared context without additional semantic justification.

364.7 Conservative inference is the default​

Unless GEAIRM or an external semantic authority explicitly states otherwise, relationships are treated as:

directed
non-symmetric
non-transitive
non-functional
without an implied inverse
without identity transfer
without ownership transfer
without Product-status transfer
without authority transfer

The exceptions are only those explicitly established by an authoritative source.

M15-R01 - Consolidation authority: M15 MUST consolidate but MUST NOT silently redefine the semantic authority or intrinsic meaning of any relationship established in M04-M14 or an external authority.

M15-R02 - Conservative inference default: All GEAIRM-owned relationships MUST follow the M03 conservative inference defaults unless M15 or the defining slice explicitly states stronger semantics.


365. Relationship Scope Categories​

M15 classifies relationship semantics into six scope categories. These categories are descriptive consolidation aids, not new ontology classes.

Scope categoryMeaningExamples
Intrinsic / classification relationRelates a semantic participant to a type/classification or capability contribution without describing a runtime occurrence.classifiedAsPrimitiveType, contributesToIntelligenceCapability
Occurrence relationDescribes what happens in a particular AI Execution, Evaluation, Decision Formation or other occurrence.involvesAIPrimitive, producesOrCausesIntelligenceEffect, evaluatesTarget
Context / role relationDescribes participation, contextual use or role in a declared context without changing intrinsic identity.includesContextResource, usesTool, groundsAIContext
Dependency / structure relationDescribes reliance, composition or coordination structure.dependsOn, participatesInComposition, participatesInFederation
Authority / governance relationDescribes bounded authority, authorization, assurance or governance-related linkage.operatesUnderDelegatedAuthority, authorizedBy, usesEvidence
Temporal / historical relationDescribes version, lifecycle, currentness, supersession or semantic change.identifiedByVersion, hasLifecycleState, supersedes

A relationship can be relevant to more than one semantic concern, but M15 avoids assigning multiple intrinsic identities to the relationship merely because it appears in several scenarios.

M15-R03 - Scope-category restraint: M15 relationship scope categories MUST NOT be treated as new ontology superclasses or used to infer semantics absent from the relationship definition.


366. Canonical Inference Contract​

M15 establishes the following normalized interpretation contract for all later GEAIRM work.

366.1 Directionality​

All GEAIRM-owned relationships are directional as defined. The reverse relationship is not inferred unless an inverse is explicitly established.

366.2 Symmetry​

No GEAIRM-owned relationship established in M04-M14 is symmetric by default.

For example:

Agent usesTool Tool-X

does not imply:

Tool-X usesTool Agent

366.3 Transitivity​

No GEAIRM-owned relationship established in M04-M14 is universally transitive.

In particular:

supersedes
withinDelegatedAuthority
participatesInComposition
participatesInFederation
participatesInOrchestration
contributesToIntelligenceCapability

are not declared transitive.

AIPS dependsOn is also not treated by GEAIRM as universally transitive; M11 represents reachability through Dependency Path.

366.4 Functional cardinality​

GEAIRM does not impose universal single-valued cardinality on its owned relationships unless explicitly established by the defining slice.

Externally governed relationships retain their authoritative cardinality. For example, the current AIPS core defines hasAgencyAutonomy and hasEmbodiment as functional Product-version properties. M15 does not alter those axioms.

366.5 No relation chaining by linguistic intuition​

GEAIRM MUST NOT infer a third relationship merely because two relationship labels appear logically composable in natural language.

For example:

Agent usesTool X
X dependsOn Y

does not automatically imply:

Agent usesTool Y

Likewise:

Execution hasAIContext C
C includesContextResource R

does not imply that every property of R applies to the Execution.

366.6 Explicit external entailments are preserved​

M15 recognizes authoritative entailments already present in external ontologies.

The current AIPS core explicitly establishes:

materiallyDependsOn
rdfs:subPropertyOf
dependsOn

and:

aips:derivedFrom
rdfs:subPropertyOf
prov:wasDerivedFrom

Those entailments are preserved because AIPS defines them.

No additional cross-relationship entailment is introduced by M15.

M15-R04 - No implicit inverse: GEAIRM MUST NOT infer inverse relationships where no inverse has been explicitly established.

M15-R05 - No implicit symmetry: GEAIRM MUST NOT infer symmetry for any M04-M14 GEAIRM-owned relationship.

M15-R06 - No implicit transitivity: GEAIRM MUST NOT infer transitivity for GEAIRM-owned relationships or for AIPS dependsOn merely from graph reachability.

M15-R07 - External entailment preservation: Explicit entailments defined by external authorities, including materiallyDependsOn -> dependsOn and aips:derivedFrom -> prov:wasDerivedFrom, MUST be preserved.

M15-R08 - No intuitive relationship chaining: Relationship chains MUST NOT create unstated semantic assertions unless an authoritative invariant explicitly permits the inference.


367. Core Mechanism and Capability Relationships​

The core mechanism/capability relationships are:

RelationshipAuthorityCanonical subjectCanonical objectScopeKey non-implications
classifiedAsPrimitiveTypeGEAIRM M05AI PrimitiveAIPS AI Primitive Type valueclassificationdoes not imply capability, effect, Product type or Agent type
contributesToIntelligenceCapabilityGEAIRM M04AI PrimitiveAIPS Intelligence Capability valuesemantic contributiondoes not imply exclusivity, sufficiency, one-to-one mapping or Product status
involvesAIPrimitiveGEAIRM M06AI ExecutionAI Primitiveoccurrencedoes not make the primitive a Product or sole causal mechanism
exercisesIntelligenceCapabilityGEAIRM M06AI ExecutionAIPS Intelligence Capability valueoccurrencedoes not imply every Product execution exercises the capability
operatesInExecutionModeGEAIRM M06AI ExecutionAIPS Execution Mode valueoccurrencedoes not alter AIPS Product-version hasExecutionMode
producesOrCausesIntelligenceEffectGEAIRM M06AI ExecutionAIPS Intelligence Effect valueoccurrencedoes not imply Productization, authorization, success or durable artifact

367.1 Primitive type, capability and effect remain independent​

The following chain is valid only when each assertion is explicit:

AI Primitive P
classifiedAsPrimitiveType PLANNER

P
contributesToIntelligenceCapability Planning

AI Execution E
involvesAIPrimitive P
exercisesIntelligenceCapability Planning
producesOrCausesIntelligenceEffect Plan

None of the later assertions is inferred solely from PLANNER.

M15-R09 - Mechanism-axis independence: AI Primitive Type, Intelligence Capability, Execution Mode and Intelligence Effect MUST remain independently asserted unless an external authority explicitly establishes a mapping.

M15-R10 - Primitive participation restraint: involvesAIPrimitive MUST NOT transfer Product, Agent, authority, capability or effect identity to the participating primitive.


368. Context, Grounding, Retrieval and Memory Relationships​

The M07 contextual relationships are:

RelationshipSubjectObjectCanonical interpretation
hasAIContextAI ExecutionAI Contextidentifies context associated with the occurrence
includesContextResourceAI Contextresource playing AI Context Resource Rolestates contextual inclusion/availability
containsInstructionPromptInstructionstates that the Prompt contains/conveys an Instruction
groundsAIContextresource playing Grounding Source RoleAI Contextstates grounding participation
retrievesResourceRetrievalresourcestates retrieval of the resource
suppliesToAIContextRetrievalAI Contextstates that retrieval supplies material to context
memoryContributesToAIContextMemoryAI Contextstates that retained state contributes to context

The model intentionally does not infer:

includesContextResource
-> groundsAIContext

retrievesResource
-> groundsAIContext

suppliesToAIContext
-> causal use

memoryContributesToAIContext
-> learning

containsInstruction
-> authorization

A resource may play multiple contextual Roles, but each Role must be independently justified.

M15-R11 - Context inclusion restraint: Inclusion or supply of a resource to AI Context MUST NOT imply grounding, trust, correctness, authority, causal use or Product status.

M15-R12 - Retrieval/grounding distinction: Retrieval relationships MUST NOT be used to infer Grounding Source Role without an explicit grounding assertion.

M15-R13 - Memory/learning distinction: memoryContributesToAIContext MUST NOT imply model learning, retraining, parameter update or Agent identity.


369. Agent, Tool and Authority Relationships​

The M08 Agent/authority relationships are:

RelationshipSubjectObjectCanonical interpretation
participatesInAIExecutionAgentAI ExecutionAgent materially participates in the occurrence
operatesUnderDelegatedAuthorityAgentDelegated AuthorityAgent operates within a bounded mandate
usesToolAgent or AI Executionresource playing Tool RoleTool is used/invoked in the declared context

External Agent Ontology properties remain separate:

hasAutonomyLevel
hasControlMode
hasToolUsePattern

and AIPS Product-version Agency / Autonomy remains separate again.

369.1 Participation is not control or accountability​

participatesInAIExecution does not imply:

  • sole execution;
  • ownership;
  • decision authority;
  • delegated authority;
  • Product identity;
  • causal responsibility for every effect;
  • organizational/legal accountability.

369.2 Tool use is not entitlement​

usesTool records use, not whether the invocation was permitted.

369.3 Authority does not imply autonomy​

operatesUnderDelegatedAuthority records scope of permission/mandate, not operational independence.

M15-R14 - Agent-participation restraint: Agent participation relationships MUST NOT imply authority, accountability, Product identity or exclusive causality.

M15-R15 - Tool-use restraint: usesTool MUST remain distinct from entitlement, authorization, ownership and AI Primitive identity.

M15-R16 - Delegation/autonomy restraint: Delegated Authority MUST remain distinct from Agent Autonomy Level, Agent Control Mode and AIPS Agency / Autonomy.


370. Environment, Control-Loop, Decision and Authorization Relationships​

The M09-M10 relationships are:

RelationshipSubjectObjectScope
occursInEnvironmentAI ExecutionEnvironmentoccurrence/environment
sensesEnvironmentAgent or AI ExecutionEnvironmentobservation/sensing
actsUponEnvironmentAgent or AI ExecutionEnvironmenteffect/interaction
participatesInControlLoopAgent or AI ExecutionControl Loopinteraction-pattern participation
participatesInDecisionFormationAgent or AI ExecutionDecision Formationdecision-process participation
informsDecisionFormationapplicable result/resource/effectDecision Formationadvisory/information contribution
authorizedBycontext-bounded execution/decision/invocation/action occurrenceAuthorizationoccurrence-specific permission
withinDelegatedAuthorityAuthorizationDelegated Authorityauthority-scope relationship

M15 preserves the following boundaries:

sensesEnvironment != Perception
actsUponEnvironment != PhysicalAction
participatesInControlLoop != Control Intelligence Capability
participatesInDecisionFormation != Decision authority
informsDecisionFormation != acceptance
authorizedBy != successful completion
withinDelegatedAuthority != permission inheritance

authorizedBy is occurrence-specific. It does not convert an Agent or Product into a permanently authorized entity.

M15-R17 - Environment/effect distinction: Environment interaction relationships MUST remain distinct from AIPS Intelligence Capability and Intelligence Effect classifications.

M15-R18 - Decision-participation restraint: Decision Formation participation or input MUST NOT imply final decision authority, acceptance, authorization or accountability.

M15-R19 - Authorization occurrence restraint: authorizedBy MUST remain a context-bounded permission relation and MUST NOT imply success, safety, low risk or unrestricted future authority.

M15-R20 - Authority nesting non-transitivity: withinDelegatedAuthority MUST NOT be treated as a transitive inheritance mechanism for all permissions.


371. Dependency, Composition, Federation and Orchestration Relationships​

M11 establishes the structural relationship family.

371.1 External dependency relationships​

aips:dependsOn
aips:materiallyDependsOn
aips:derivedFrom

remain externally governed.

materiallyDependsOn is a subproperty of dependsOn, but materiality itself is not treated as transitive.

371.2 GEAIRM structural participation relationships​

RelationshipSubjectObjectCanonical interpretation
participatesInCompositionapplicable participantCompositionparticipant intentionally combined/coordinated in the arrangement
participatesInFederationindependently governed participantFederationparticipant cooperates while retaining identity/governance
participatesInOrchestrationapplicable participantOrchestrationparticipant coordinates or is coordinated

These relationships do not infer:

Composition -> Product
Composition -> Compound Product
Composition -> AIPS Composite

Federation -> CrossDomainFederated
Federation -> Collective Agent

Orchestration -> EnterpriseOrchestrated
Orchestration -> Artificial Intelligence
Orchestration -> AI Product

371.3 Dependency reachability is path semantics​

A direct dependsOn edge and a multi-edge Dependency Path remain distinct. GEAIRM does not infer a direct dependency edge merely because reachability exists through intermediates.

M15-R21 - Dependency/path distinction: Direct dependency assertions MUST remain distinct from transitive reachability represented through Dependency Path.

M15-R22 - Material-dependency restraint: materiallyDependsOn MUST NOT be treated as universally transitive or as automatic risk/governance inheritance.

M15-R23 - Structural participation non-productization: Composition, Federation and Orchestration participation MUST NOT create Product, AI Product, Compound Product, Collective Agent or AIPS topology-scope identity by inference.


372. Evaluation and Assurance Relationships​

M13 establishes the assessment relationship family.

RelationshipSubjectObjectCanonical interpretation
evaluatesTargetEvaluationparticipant/property playing Evaluation Target Roleidentifies what is assessed
usesEvaluationCriterionEvaluationEvaluation Criterionidentifies criterion applied
producesEvaluationResultEvaluationEvaluation Resultidentifies result of assessment
recordedInEvaluationReportEvaluation ResultAIPS AI Evaluation Reportidentifies report recording/summarizing result
usesEvidenceEvaluation or Assurance Assessmentresource playing Evidence Roleidentifies evidence used
assessesAssuranceOfAssurance Assessmentparticipant/property/claimidentifies assurance concern target
producesAssuranceFindingAssurance AssessmentAssurance Findingidentifies assurance conclusion

The relationship graph does not imply:

Evaluation Result -> Assurance Finding
Evidence -> proof
AIEvalReport -> PASS
Assurance Finding -> Risk Tier
Evaluation -> Product qualification

Any such mapping requires the applicable AIPS or external governance rule.

M15-R24 - Evaluation/assurance separation: Evaluation relationships and Assurance relationships MUST remain distinct even where Evaluation Results are used as Evidence in Assurance.

M15-R25 - Evidence sufficiency restraint: usesEvidence MUST NOT imply that the evidence is sufficient, accepted, current, authoritative or supportive.

M15-R26 - Assurance/risk restraint: producesAssuranceFinding MUST NOT directly determine AIPS Effective Risk Tier absent authoritative AIPS governance computation.


373. Identity, Version, Lifecycle and Currentness Relationships​

M14 establishes the temporal/historical relationship family.

RelationshipSubjectObjectCanonical interpretation
identifiedByVersionresource playing Versioned Resource RoleVersion Identifieridentifies represented revision
hasLifecycleStategoverned participantLifecycle Staterecords lifecycle standing under authority/time context
assessesCurrentnessOfCurrentness Assessmenttargetidentifies target of currentness assessment
hasCurrentnessConcernCurrentness Assessmentconcern/categoryidentifies what aspect of currentness is assessed
hasCurrentnessStatusCurrentness Assessmentstatus under authorityrecords resulting currentness status
supersedeslater governed resourceearlier governed resourcereplacement for declared current purpose while preserving history
affectsSemanticTargetSemantic Change Eventgoverned participant/versionidentifies target potentially requiring reassessment

AIPS hasVersion remains Product-specific.

The temporal model does not infer:

same Version Identifier -> same identity
supersedes -> equivalent
supersedes -> compatible
supersedes -> deletion
newer version -> downstream adoption
affectsSemanticTarget -> actual semantic change outcome
Lifecycle State -> Currentness

M15-R27 - Version/identity restraint: Version identifiers and succession relationships MUST NOT substitute for independent identity semantics.

M15-R28 - Supersession non-equivalence: supersedes MUST NOT imply semantic equivalence, backward compatibility, deletion or historical mutation.

M15-R29 - Change-event restraint: affectsSemanticTarget MUST identify potential impact only and MUST NOT pre-decide reclassification, currentness status, lifecycle transition or version increment.


374. Externally Governed Relationship Bridge​

The following externally governed relationship families are explicitly integrated but not owned by GEAIRM.

Relationship/propertyAuthorityGoverned scope retained by M15
aips:hasVersionAIPSProductIdentity -> AIProductVersion
aips:hasIntelligenceCapabilityAIPSAIProductVersion -> IntelligenceCapability
aips:hasExecutionModeAIPSAIProductVersion -> ExecutionMode
aips:hasIntelligenceEffectAIPSAIProductVersion -> IntelligenceEffect
aips:hasAgencyAutonomyAIPSAIProductVersion -> AgencyAutonomy; functional in current AIPS core
aips:hasEmbodimentAIPSAIProductVersion -> Embodiment; functional in current AIPS core
aips:hasTopologyScopeAIPSAIPS Product scope as defined by AIPS
aips:hasLifecycleStageAIPSAIPS Product lifecycle scope; functional in current AIPS core
aips:dependsOnAIPSgeneral AIPS dependency relation
aips:materiallyDependsOnAIPSmaterial dependency; subproperty of dependsOn
aips:derivedFromAIPS / PROVprovenance derivation; subproperty of prov:wasDerivedFrom
aips:dependencyPathAIPSAIPrimitiveEvidence qualification path
agent:hasAutonomyLevelKivaNura Agent OntologyAgent -> Agent Autonomy Level
agent:hasControlModeKivaNura Agent OntologyAgent -> Agent Control Mode
agent:hasToolUsePatternKivaNura Agent OntologyAgent -> Tool Use Pattern

M15 explicitly prohibits using the GEAIRM generic relationships to broaden these Product/Agent properties.

For example:

AI Execution operatesInExecutionMode Interactive

does not create:

AI Execution aips:hasExecutionMode Interactive

because the AIPS property is Product-version scoped.

M15-R30 - External property domain preservation: GEAIRM MUST NOT broaden externally governed relationship domains/ranges merely to simplify integrated modelling.

M15-R31 - External cardinality preservation: Functional or other cardinality semantics explicitly defined by an external authority MUST remain authoritative in that authority's scope.


375. Relationship-Level Identity Preservation​

M15 consolidates a general invariant already repeated throughout M04-M14:

A relationship assertion changes the graph of what is known about participants; it does not, by itself, change their intrinsic identity.

Examples:

AI Product plays Tool Role
-> still AI Product

Knowledge Resource groundsAIContext
-> still Knowledge Resource

Agent participatesInComposition
-> still Agent

AI Primitive participates in Product constitution
-> still AI Primitive

Evaluation Result plays Evidence Role
-> still Evaluation Result

The exception is only where an authoritative relationship is itself part of the criteria that establish another classification and that classification is explicitly evaluated. For example, AIPS may use material upstream AI Product dependencies as evidence in determining Derived APT. The graph edge does not perform the classification by itself; the authoritative AIPS rule does.

M15-R32 - Identity-preserving relationship rule: GEAIRM relationship participation MUST preserve participant identity unless an explicit external or GEAIRM classification rule independently establishes a new classification.


376. Relationship-Level Authority Preservation​

A relationship does not transfer semantic authority from one endpoint to another.

For example:

GEAIRM Agent usesTool AIPS AI Product

does not make GEAIRM authoritative for AI Product semantics.

Likewise:

GEAIRM AI Execution involvesAIPrimitive AIPS AI Primitive

does not transfer the definition of AI Primitive from AIPS to GEAIRM.

The same rule applies to:

  • BPS Product concepts;
  • Agent Ontology concepts;
  • PROV entities/derivations;
  • GEDRM Data concepts;
  • GEARM Information/Analytics concepts;
  • external policy/governance concepts.

M15-R33 - Authority non-transfer: Semantic authority MUST NOT transfer through dependency, participation, context, composition, evidence, version or other relationships.


377. Canonical Hard-Distinction Catalogue​

M15 consolidates the hard distinctions into twelve semantic clusters.

377.1 Metamodel distinctions​

Concept != Classification
Concept != Role
Concept != State
Classification != Role
Classification != State
Role != intrinsic type
State != taxonomic specialization
label similarity != identity
semantic mapping != identity unless explicitly equivalent

377.2 Core AI distinctions​

Artificial Intelligence != AI Product
Artificial Intelligence != AI Primitive
Artificial Intelligence != AI Model
Artificial Intelligence != Agent
Artificial Intelligence != automation

Intelligence Capability != AI Primitive
AI Primitive Type != Intelligence Capability
AI Primitive Type != Intelligence Effect
AI Primitive Type != Execution Mode

377.3 Mechanism / execution / effect distinctions​

Learned Model primitive != AIModel != Model Product
Planner primitive != Planning capability != Plan effect
Optimizer primitive != Optimization capability
Control capability != Control Loop != governance Control

Execution Mode != capability
Execution Mode != effect
Execution Mode != Agent identity
Execution Mode != deployment architecture

377.4 Context / grounding distinctions​

AI Context != M03 Modelling Context
AI Context != BPS Business Context
AI Context != Environment
AI Context != Memory
AI Context != Prompt

Prompt != Instruction
Instruction != authorization
Rule != Symbolic Reasoner
Knowledge Resource != Knowledge Reasoner

Retrieval != Grounding
Retrieval != Intelligent Search
Grounding != correctness
Grounding Source Role != intrinsic resource type

377.5 Agent / autonomy / authority distinctions​

Agent != AI Primitive
Agent != AI Product
Agent != AgenticGoalDriven Execution Mode

AIPS Agency/Autonomy != Agent Autonomy Level
Agent Autonomy Level != Agent Control Mode
Agent Autonomy Level != Delegated Authority
Delegated Authority != Authorization

HumanInLoop != ActWithApproval
FullyAutomated != AutonomousAction
DynamicToolUse != unrestricted Tool authority

Tool Role != AI Primitive
Tool Role != Product Family
Tool use != authorization

377.6 Environment / embodiment distinctions​

AIPS Embodiment != Environment Kind
AIPS Embodiment != RoboticAgent
AIPS Embodiment != Agent Substrate Type

Sensing != Perception
Actuation != Control capability
Actuation != PhysicalAction

Control Loop != ClosedLoopAutonomous
PhysicalAction != EmbodiedActuating automatically
physical capability != permission

377.7 Decision / action distinctions​

RecommendationEffect != Decision
RecommendationEffect != AIPS Recommend AR
Plan != Decision
Plan != Authorization
Decision effect != Decision authority
Decision effect != Authorization
Decision effect != Action

Authorization != Delegated Authority
ActWithApproval != Authorization
AutonomousAction != unrestricted authority
Action effect != successful outcome

377.8 Dependency / composition / topology distinctions​

dependsOn != materiallyDependsOn
direct dependency != Dependency Path reachability
derivedFrom != dependsOn
material dependency != universal criticality

Composition != Product
Composition != Compound Product
Composition != AIPS Composite
Composition != Hybrid Intelligence Engine
Composition != Collective Agent

Federation != CrossDomainFederated
Orchestration != EnterpriseOrchestrated
AIPS Topology Scope != deployment topology

377.9 Product-boundary distinctions​

Productness != AI Product qualification
AI Primitive != Product
AI Primitive != AI Product
AI Asset != Product
Model Product != AI Product
AI-enabled Product != AI Product automatically
Agent != Product
Intelligence Effect != Product
APT applies only after AI Product qualification
GEAIRM conformance != AIPS qualification

377.10 Evaluation / assurance / risk distinctions​

Evaluation != AI Evaluation Report
Evaluation != Assurance Assessment
Evaluation Result != Assurance Finding
Evidence != proof
Evidence != Assurance
Evidence Role != ExplanationEvidence

Assurance Finding != Effective Risk Tier
Fitness for Purpose != Effective Risk Tier
PASS assurance != low risk
Effective Risk Tier != APT

377.11 Governance distinctions​

Policy != Rule
Policy != Instruction
Policy != Governance Obligation
Governance Obligation != Governance Control
Governance Obligation != concrete authorization occurrence

AIPS Control capability
!= Agent ControlMode
!= GEAIRM Control Loop
!= governance Control

governance evidence != proof of compliance
dependency/composition != governance inheritance

377.12 Identity / temporal distinctions​

identity != label
identity != classification
Product Identity != AI Product Version
Version Identifier != identity
Lifecycle State != Currentness
Currentness != freshness
Currentness != availability
Currentness != Fitness for Purpose

supersession != deletion
supersession != equivalence
Provenance != Dependency
Provenance != Evidence Role
Lineage != one generic edge type

new upstream version != automatic downstream adoption
currentness change != historical mutation

These distinctions are semantic invariants. They are not automatically OWL disjointness axioms.

M15-R34 - Hard-distinction preservation: Later GEAIRM slices, examples and executable artifacts MUST preserve the M15 hard-distinction catalogue unless an explicit normative revision changes the model.

M15-R35 - Non-equivalence/disjointness restraint: M15 hard distinctions MUST NOT automatically be encoded as formal disjointness where the intended requirement is only non-equivalence or non-inference.


378. Cross-Slice Invariant Catalogue​

The following invariant identifiers are human-readable GEAIRM specification identifiers. They are not executable namespace IRIs; executable formalization remains deferred to A3-A8.

InvariantCanonical requirement
M15-I01 Authority PreservationReuse does not transfer semantic authority to GEAIRM.
M15-I02 Identity PreservationParticipation in a relationship does not by itself change intrinsic identity.
M15-I03 Product Non-InferenceAI participation, dependency, effect, Tool use or composition does not by itself create Product identity.
M15-I04 AI Product Non-InferenceAI-related participation does not bypass BPS Productness and AIPS AQ1-AQ5.
M15-I05 Role Non-IntrinsicalityContextual Roles do not become intrinsic types by participation.
M15-I06 State Non-TaxonomyState does not become taxonomic specialization by transition.
M15-I07 Classification OrthogonalityPrimitive Type, Capability, Mode, Effect, AR, Embodiment, Topology and Risk remain distinct axes.
M15-I08 Occurrence/Product SeparationOccurrence assertions do not automatically become Product-Version declarations.
M15-I09 Product-Property ScopeAIPS Product properties retain their authoritative domain/range.
M15-I10 Context QualificationContext-bounded assertions are not generalized beyond their scope automatically.
M15-I11 No Lexical EquivalenceShared labels do not establish identity/equivalence.
M15-I12 No Dependency InheritanceDependency does not transfer ontology identity, authority, risk or governance.
M15-I13 Path not TransitivityMulti-edge reachability is represented through Dependency Path, not universal transitivity.
M15-I14 Materiality Non-TransitivityMaterial dependency is not universally transitive.
M15-I15 Composition Non-InheritanceComposition does not transfer participant identity or Product Family.
M15-I16 Runtime Composition Non-ProductizationRuntime orchestration/composition does not automatically create a new AI Product.
M15-I17 Agent/AI SeparationAgent identity does not imply AI Primitive, AI Product or AgenticGoalDriven mode.
M15-I18 Autonomy/Authority SeparationOperational autonomy, control mode, AR, Delegated Authority and Authorization remain separate.
M15-I19 Tool/Authority SeparationTool availability/use does not establish entitlement or Authorization.
M15-I20 Instruction/Authority SeparationInstruction does not confer permission.
M15-I21 Sensing/Perception SeparationSensing does not imply Perception Intelligence Capability.
M15-I22 Actuation/Action SeparationActuation does not equal PhysicalAction Intelligence Effect.
M15-I23 Decision/Authority SeparationDecision effect does not imply decision authority or Authorization.
M15-I24 No Mandatory Decision PipelineRecommendation, Plan, Decision, Authorization and Action are not a universal sequence.
M15-I25 Effect Non-ProductizationIntelligence Effects do not automatically become Products or durable assets.
M15-I26 Evaluation/Assurance SeparationEvaluation and Assurance Assessment remain distinct.
M15-I27 Evidence Non-ProofEvidence presence does not prove correctness, sufficiency or compliance.
M15-I28 Assurance/Risk OrthogonalityAssurance outcome does not directly determine Effective Risk Tier.
M15-I29 Risk/Governance AuthorityAIPS remains authority for R0-R4, Risk Driver Assessment and Governance Obligation Profile.
M15-I30 Obligation/Control SeparationGovernance obligation is not the control mechanism or concrete occurrence satisfying it.
M15-I31 Identity/Version SeparationContinuing identity and version-specific semantic state remain distinct.
M15-I32 Historical Non-MutationLater change/currentness does not rewrite historical version semantics.
M15-I33 Currentness ContextualityCurrentness is target-, concern-, time- and authority-qualified.
M15-I34 Supersession History PreservationSupersession replaces for current purpose without deleting history.
M15-I35 Provenance/Dependency SeparationDerivation provenance does not automatically establish current dependency.
M15-I36 Architecture BoundaryGEAIRM semantics do not prescribe GEAIRA planes, ABBs, registries or technology topology.

These invariants are the primary input to the later A3 Semantic Invariant Catalogue. A3 may classify individual invariants as machine-executable, hybrid-assessment or assessment-only; M15 does not pre-assign that executable status.

M15-R36 - Invariant authority: The M15 invariant catalogue MUST be treated as normative consolidation of M01-M14 rather than optional explanatory guidance.

M15-R37 - Executability deferral: M15 MUST NOT claim that every invariant is machine-executable; executable/hybrid/assessment-only treatment remains part of the A3-A14 formalization programme.


379. Canonical GEAIRM-Owned Relationship Inventory​

The following inventory is the consolidated set of GEAIRM-owned relationship types established in M04-M14.

RelationshipIntroducedSubject expectationObject expectationPrimary scope
contributesToIntelligenceCapabilityM04AI PrimitiveIntelligence Capabilitymechanism/capability
classifiedAsPrimitiveTypeM05AI PrimitiveAI Primitive Typeclassification
involvesAIPrimitiveM06AI ExecutionAI Primitiveoccurrence
exercisesIntelligenceCapabilityM06AI ExecutionIntelligence Capabilityoccurrence
operatesInExecutionModeM06AI ExecutionExecution Modeoccurrence
producesOrCausesIntelligenceEffectM06AI ExecutionIntelligence Effectoccurrence
hasAIContextM07AI ExecutionAI Contextoccurrence/context
includesContextResourceM07AI Contextcontextual resourcecontext
containsInstructionM07PromptInstructioncontent/context
groundsAIContextM07resource playing Grounding Source RoleAI Contextgrounding
retrievesResourceM07Retrievalresourceretrieval
suppliesToAIContextM07RetrievalAI Contextcontext supply
memoryContributesToAIContextM07MemoryAI Contextmemory/context
participatesInAIExecutionM08AgentAI ExecutionAgent participation
operatesUnderDelegatedAuthorityM08AgentDelegated Authorityauthority
usesToolM08Agent or AI Executionresource playing Tool RoleTool use
occursInEnvironmentM09AI ExecutionEnvironmentenvironment
sensesEnvironmentM09Agent or AI ExecutionEnvironmentsensing
actsUponEnvironmentM09Agent or AI ExecutionEnvironmentenvironment action
participatesInControlLoopM09Agent or AI ExecutionControl Loopfeedback/control
participatesInDecisionFormationM10Agent or AI ExecutionDecision Formationdecision process
informsDecisionFormationM10applicable resource/result/effectDecision Formationdecision input
authorizedByM10context-bounded occurrenceAuthorizationauthorization
withinDelegatedAuthorityM10AuthorizationDelegated Authorityauthority scope
participatesInCompositionM11applicable participantCompositionstructure
participatesInFederationM11independently governed participantFederationcoordination
participatesInOrchestrationM11applicable participantOrchestrationcoordination
evaluatesTargetM13EvaluationEvaluation Target-role participant/propertyevaluation
usesEvaluationCriterionM13EvaluationEvaluation Criterionevaluation
producesEvaluationResultM13EvaluationEvaluation Resultevaluation
recordedInEvaluationReportM13Evaluation ResultAIPS AI Evaluation Reportevidence/reporting
usesEvidenceM13Evaluation or Assurance Assessmentresource playing Evidence Roleevidence
assessesAssuranceOfM13Assurance Assessmentparticipant/property/claimassurance
producesAssuranceFindingM13Assurance AssessmentAssurance Findingassurance
identifiedByVersionM14Versioned Resource-role participantVersion Identifierversion
hasLifecycleStateM14governed participantLifecycle Statelifecycle
assessesCurrentnessOfM14Currentness Assessmenttargetcurrentness
hasCurrentnessConcernM14Currentness Assessmentconcern/categorycurrentness
hasCurrentnessStatusM14Currentness Assessmentauthority-specific statuscurrentness
supersedesM14later governed resourceearlier governed resourcehistorical succession
affectsSemanticTargetM14Semantic Change Eventgoverned participant/versionchange impact

M15 introduces no additional GEAIRM-owned relationship type.

M15-R38 - Inventory completeness: Later GEAIRM 0.1 work MUST treat the M15 relationship inventory as the canonical consolidated M04-M14 relationship baseline unless a later normative slice explicitly amends it.


380. Relationship Anti-Patterns​

The following modelling practices are prohibited unless an explicit authoritative rule justifies them.

380.1 Shadow external property​

geairm:hasProductVersion

created merely to duplicate aips:hasVersion.

380.2 Occurrence-to-Product promotion​

Execution E produces PhysicalAction
therefore Product P hasEmbodiment ClosedLoopAutonomous

without independent AIPS Product classification.

380.3 Context-role type promotion​

Resource R groundsAIContext C
therefore R rdf:type GroundingResource

when M07 defines Grounding Source as a contextual Role.

380.4 Dependency closure as direct edge​

A dependsOn B
B dependsOn C
therefore assert A dependsOn C

without preserving that the conclusion is path reachability rather than a direct dependency assertion.

380.5 Authority leakage​

Agent usesTool Product-P
therefore Agent inherits P's ActionAuthorization

380.6 Product leakage​

Agent + Primitive + Tools
therefore AI Product

without BPS/AIPS qualification.

380.7 Risk leakage​

Upstream Product is R4
therefore downstream Composition is R4

without AIPS risk computation.

380.8 Historical overwrite​

v2 changes classification
therefore rewrite v1 classification

380.9 Lexical equivalence​

HumanInLoop == ActWithApproval
Control == ControlMode
Composite == Composition

based on labels alone.

380.10 Architecture leakage​

usesTool
therefore Tool Registry ABB

hasAIContext
therefore Context Service

usesEvidence
therefore Evidence Store

M15 relationship semantics do not prescribe those architecture decisions.

M15-R39 - Anti-pattern prohibition: GEAIRM examples, profiles and machine artifacts MUST NOT encode M15 anti-patterns as normative semantics.


381. Canonical Relationship Validation Questions​

Any future GEAIRM assertion or profile SHOULD be reviewable through the following questions:

  1. Authority: Who owns the relationship and endpoint concepts?
  2. Identity: Does the assertion preserve endpoint identity?
  3. Scope: Is the relationship intrinsic, contextual, occurrence-level, Product-scoped or temporal?
  4. Context: Is a contextual assertion being generalized beyond its declared context?
  5. Cardinality: Is an unproven single-valued assumption being introduced?
  6. Inference: Is symmetry, transitivity, inverse or chaining being assumed without authority?
  7. Product boundary: Does the assertion accidentally create Product/AI Product status?
  8. Authority boundary: Does the assertion accidentally grant permission, delegation or accountability?
  9. Temporal boundary: Does a current-state assertion rewrite historical truth?
  10. Architecture boundary: Is semantic meaning being converted into a prescribed implementation component?

These questions are normative review guidance but are not themselves a new conformance algorithm.

M15-R40 - Relationship review discipline: Profiles and examples SHOULD apply the M15 relationship validation questions before introducing any new relationship mapping or inference.


382. M15 Normative Rules​

RuleRequirement
M15-R01M15 MUST consolidate without silently redefining semantic authority or intrinsic meaning from M04-M14.
M15-R02GEAIRM-owned relationships MUST follow M03 conservative inference defaults unless explicitly strengthened.
M15-R03M15 relationship scope categories MUST NOT become new ontology classes or inference shortcuts.
M15-R04Inverse relationships MUST NOT be inferred where not explicitly established.
M15-R05Symmetry MUST NOT be inferred for M04-M14 GEAIRM-owned relationships.
M15-R06Transitivity MUST NOT be inferred for GEAIRM-owned relationships or AIPS dependsOn merely from graph reachability.
M15-R07Explicit external entailments, including material dependency and PROV derivation subproperty semantics, MUST be preserved.
M15-R08Relationship chaining MUST NOT create unstated assertions absent explicit authoritative inference rules.
M15-R09AI Primitive Type, Intelligence Capability, Execution Mode and Intelligence Effect MUST remain independent axes.
M15-R10AI Primitive participation MUST NOT transfer Product, Agent, authority, capability or effect identity.
M15-R11Context inclusion/supply MUST NOT imply grounding, trust, correctness, authority, causal use or Product status.
M15-R12Retrieval relationships MUST NOT imply Grounding Source Role without explicit grounding semantics.
M15-R13Memory contribution to context MUST NOT imply learning, retraining, parameter update or Agent identity.
M15-R14Agent participation MUST NOT imply authority, accountability, Product identity or exclusive causality.
M15-R15Tool use MUST remain distinct from entitlement, authorization, ownership and AI Primitive identity.
M15-R16Delegated Authority MUST remain distinct from Agent autonomy, Control Mode and AIPS Agency/Autonomy.
M15-R17Environment interaction relationships MUST remain distinct from Intelligence Capability and Intelligence Effect classifications.
M15-R18Decision Formation participation/input MUST NOT imply final decision authority, acceptance, authorization or accountability.
M15-R19authorizedBy MUST remain context-bounded and MUST NOT imply success, safety, low risk or unrestricted future authority.
M15-R20withinDelegatedAuthority MUST NOT be treated as a transitive permission-inheritance relation.
M15-R21Direct dependency assertions MUST remain distinct from Dependency Path reachability.
M15-R22materiallyDependsOn MUST NOT be treated as universally transitive or as automatic risk/governance inheritance.
M15-R23Composition/Federation/Orchestration participation MUST NOT create Product, AI Product, Compound Product, Collective Agent or AIPS topology-scope identity.
M15-R24Evaluation relationships and Assurance relationships MUST remain distinct.
M15-R25usesEvidence MUST NOT imply evidence sufficiency, acceptance, currentness, authority or supportiveness.
M15-R26Assurance Finding MUST NOT directly determine AIPS Effective Risk Tier absent AIPS governance computation.
M15-R27Version identifiers and succession relationships MUST NOT substitute for independent identity semantics.
M15-R28Supersession MUST NOT imply semantic equivalence, backward compatibility, deletion or historical mutation.
M15-R29Semantic change targeting MUST NOT pre-decide reclassification, currentness status, lifecycle transition or version increment.
M15-R30GEAIRM MUST NOT broaden externally governed property domains/ranges for integrated modelling convenience.
M15-R31External cardinality semantics explicitly defined by an authority MUST remain authoritative in that scope.
M15-R32Relationship participation MUST preserve participant identity unless an independent authoritative classification rule establishes otherwise.
M15-R33Semantic authority MUST NOT transfer through relationships.
M15-R34Later slices, examples and executable artifacts MUST preserve the M15 hard-distinction catalogue unless normatively revised.
M15-R35Non-equivalence MUST NOT automatically be encoded as formal OWL disjointness.
M15-R36The M15 invariant catalogue is normative consolidation of M01-M14.
M15-R37M15 MUST NOT claim every invariant is machine-executable; executability classification is deferred.
M15-R38The M15 relationship inventory is the canonical consolidated M04-M14 relationship baseline unless normatively amended.
M15-R39GEAIRM examples, profiles and machine artifacts MUST NOT encode the M15 relationship anti-patterns as normative semantics.
M15-R40Profiles/examples SHOULD apply the M15 relationship validation questions before introducing new relationship mappings or inference.
M15-R41Occurrence-level relationships MUST NOT automatically populate Product-Version declarations.
M15-R42Product-Version declarations MUST NOT be treated as assertions that every runtime occurrence exhibits every declared capability, mode or effect.
M15-R43Contextual Roles MUST NOT become intrinsic endpoint types solely because a relationship references the Role.
M15-R44Relationship assertions MUST NOT silently create ownership, accountability, entitlement or permission semantics.
M15-R45Relationship assertions MUST NOT silently create architecture components, registries, services, runtime topology or deployment requirements.
M15-R46Historical/temporal relationships MUST preserve M14 historical non-mutation semantics.
M15-R47GEAIRM relationship mappings to GEDRM, GEARM, BPS, AIPS, Agent Ontology or PROV MUST preserve the source authority's semantic scope.
M15-R48Future executable relationship constraints MUST preserve authorized external entailments while preventing unauthorized cross-slice inference.
M15-R49M15 introduces no new Product, Agent, risk, lifecycle or architecture taxonomy.
M15-R50M15 semantic consolidation MUST remain technology-neutral and implementation-independent.

383. M15 Exit Criteria​

GEAIRM-M15 is complete when:

  • all GEAIRM-owned relationship types established in M04-M14 are consolidated into one canonical inventory;
  • relationship ownership and external authority boundaries are explicit;
  • subject/object expectations are normalized;
  • occurrence, context, dependency, authority and temporal relationship scopes are distinguished;
  • conservative directionality, symmetry, transitivity, inverse and cardinality defaults are explicit;
  • AIPS materiallyDependsOn -> dependsOn is preserved as an authorized external entailment;
  • AIPS derivedFrom -> prov:wasDerivedFrom is preserved as an authorized external entailment;
  • no new unauthorized relationship chaining is introduced;
  • Product-scoped AIPS properties remain Product-scoped;
  • Agent Ontology autonomy/control/tool-use properties remain Agent-scoped;
  • contextual-role relationships preserve endpoint identity;
  • occurrence assertions remain distinct from Product declarations;
  • dependency paths remain distinct from direct dependency edges;
  • Composition, Federation and Orchestration remain non-productizing by themselves;
  • Evaluation, Evidence and Assurance relationships remain distinct;
  • version/currentness/supersession relationships preserve historical truth;
  • semantic authority does not transfer through relationships;
  • the twelve-cluster hard-distinction catalogue is consolidated;
  • the M15-I01 through M15-I36 invariant catalogue is established;
  • the invariant catalogue remains separate from executable formalization status;
  • relationship anti-patterns are explicit;
  • no new GEAIRM domain concept or relationship type is introduced by the consolidation itself;
  • no GEAIRA architecture is prescribed;
  • M01-M14 semantic authority, abstraction and historical-preservation rules remain intact.

M15 conclusion: these criteria are satisfied by this slice. GEAIRM now has a single canonical relationship and invariant baseline that preserves relationship authority, endpoint identity, contextual scope, Product boundaries, historical truth and conservative inference across the complete M04-M14 semantic model.


384. Forward Work​

The next slice is:

GEAIRM-M16 - Canonical Patterns, Examples & Counterexamples

M16 will exercise the M15 relationship/invariant baseline through representative enterprise-AI patterns and deliberately invalid counterexamples. It will cover model-based AI, symbolic reasoning, RAG/grounded generation, Agent/tool use, human-in-the-loop decisions, autonomous digital action, Physical AI, Derived/Composite AI Products, multi-Agent composition, Evaluation/Assurance, currentness change and historical reclassification. M16 will validate semantic expressiveness without adding architecture or silently weakening M15 invariants.


385. GEAIRM-M16 - Canonical Patterns, Examples & Counterexamples​

385.1 Purpose of this slice​

M15 consolidated the canonical GEAIRM relationship model and the M15-I01 through M15-I36 invariant catalogue. M16 now stress-tests that semantic model through representative enterprise-AI patterns, valid examples and deliberately invalid counterexamples.

M16 answers six questions:

  1. Can GEAIRM represent learned, symbolic, grounded, agentic and embodied AI without collapsing them into one model-centric ontology?
  2. Can the model distinguish AI mechanisms, capabilities, execution, effects, Agents, Products, authority and governance in realistic scenarios?
  3. Can AIPS Product qualification and Product-version semantics be integrated without turning every GEAIRM participant into an AI Product?
  4. Can dependency, composition, Federation and Orchestration be expressed without graph-shape ontology inheritance?
  5. Can Evaluation, Assurance, Risk, Currentness and historical reclassification remain distinct in end-to-end examples?
  6. Do the M15 hard distinctions survive deliberately tempting counterexamples?

M16 introduces no new GEAIRM-owned concept, relationship, Product type, Agent type, risk tier, lifecycle vocabulary or architecture construct.

The patterns are canonical interpretation examples. They test and illustrate M04-M15 semantics; they do not override those semantics.

M16-R01 - Example subordination: M16 patterns and counterexamples MUST be interpreted under M01-M15 normative semantics and MUST NOT silently redefine them.

M16-R02 - No semantic expansion by example: An entity, label or connection appearing only in an M16 example MUST NOT become a new canonical GEAIRM concept or relationship merely by appearing in the example.


386. Pattern Record and Validation Method​

Each canonical M16 pattern uses the following record:

FieldMeaning
Pattern IDStable human-readable M16 pattern identifier
IntentSemantic problem the pattern demonstrates
ParticipantsConcepts/roles/classifications involved
Canonical assertionsValid relationships/classifications under existing authority
What may be concludedAuthorized interpretation
What must not be concludedProhibited inference
CounterexampleDeliberately invalid interpretation
Invariants exercisedM15-Ixx constraints stressed by the pattern

The identifiers in this slice are documentation identifiers, not RDF/OWL IRIs. Stable machine identifiers remain deferred to A1-A8.

A canonical pattern may use external AIPS or Agent-Ontology properties where those authorities own the semantics. Such use does not transfer authority to GEAIRM.

M16-R03 - Canonical relationship use: M16 canonical assertions MUST use relationships from the M15 canonical inventory or externally authoritative relationships already bridged by M15.

M16-R04 - Counterexample semantics: An M16 counterexample is invalid because it violates existing GEAIRM/AIPS/BPS/Agent-Ontology semantics; failure of a counterexample MUST NOT be interpreted as creation of a new rule solely by example.


387. Scenario Coverage Matrix​

PatternPrimary semantic concerns exercised
M16-P01Learned-model AI, Model Product boundary, Primary APT
M16-P02Model Product as Product + AI Primitive but not AI Product
M16-P03Symbolic reasoning, Rules/Knowledge, no mandatory ML
M16-P04Retrieval, Grounding, AI Context, generated effect, RAG boundary
M16-P05Agent participation, Tool Role, delegated authority
M16-P06Recommendation, human Decision Formation, Authorization, action
M16-P07Autonomous digital action under bounded authority
M16-P08Physical AI, sensing, control loop, PhysicalAction, Embodiment
M16-P09Derived AI Product, Composite derivation topology
M16-P10Runtime orchestration of AI Products without new Product identity
M16-P11Multi-Agent Composition without Collective Agent/Product inference
M16-P12AI-enabled Service Product and PIN boundary
M16-P13Evaluation, Evidence, Assurance, Risk separation
M16-P14Currentness degradation, pinned dependencies, historical preservation
M16-P15AI Primitive removal and Product-family reclassification across versions

Together the patterns exercise the major concepts introduced in M04-M14 and the M15 invariant catalogue.


388. M16-P01 - Primary Learned-Model Fraud Detection AI Product​

Intent​

Demonstrate a conventional learned-model-backed AI Product without treating the model itself as the Productized intelligence proposition.

Participants​

  • AIPS AI Product Version: Fraud Detection Product v2.1
  • AIPS Model Product: Fraud Model v4
  • AI Primitive: the governed learned-model primitive
  • Data Products: payment lifecycle and fraud-history data
  • Intelligence Capabilities: Detection, Classification, Prediction
  • Intelligence Effects: Score / ClassificationEffect / DetectionAlert as applicable
  • APT: Primary

Canonical assertions​

FraudModel-v4
classifiedAsPrimitiveType LEARNED_MODEL

FraudModel-v4
contributesToIntelligenceCapability Detection
contributesToIntelligenceCapability Classification
contributesToIntelligenceCapability Prediction

FraudProduct-v2.1
aips:dependsOn FraudModel-v4
aips:dependsOn PaymentLifecycleDataProduct
aips:dependsOn FraudHistoryDataProduct

FraudProduct-v2.1
AIPS AI Product qualification = CURRENT
APT = Primary

A runtime occurrence may separately state:

AI Execution E1
involvesAIPrimitive FraudModel-v4
exercisesIntelligenceCapability Detection
operatesInExecutionMode RealTimeEventDriven
producesOrCausesIntelligenceEffect DetectionAlert

Authorized conclusion​

The Product may be Primary even though it depends on Data Products and a Model Product, because AIPS Derived classification depends on material upstream AI Product intelligence provenance rather than ordinary Model Product/Data Product dependencies.

Prohibited conclusions​

FraudModel-v4 is a Model Product
-> therefore FraudProduct-v2.1 is Derived [INVALID]

E1 produces DetectionAlert
-> therefore every FraudProduct execution
produces DetectionAlert [INVALID]

M16-C01 - Counterexample​

FraudProduct-v2.1 materiallyDependsOn FraudModel-v4
therefore
APT = Derived

This is invalid because a Model Product / AI Primitive dependency does not by itself establish a material upstream AI Product dependency.

Invariants exercised​

M15-I04, M15-I07, M15-I08, M15-I09, M15-I12, M15-I25.


389. M16-P02 - Productized Model Boundary​

Intent​

Demonstrate that a governed downloadable model may simultaneously be a Product and AI Primitive without automatically qualifying as an AIPS AI Product.

Participants​

  • BPS Product
  • AIPS Model Product
  • AI Primitive
  • contextual Business Product role for the vendor
  • downstream AI Product dependency

Canonical assertions​

QwenLikeModel-v7
qualifies as Product under BPS
Product Kind = Model Product
may play Business Product role for VendorCo
classifiedAsPrimitiveType LEARNED_MODEL

A downstream Product may assert:

ResearchAssistant-v1
aips:dependsOn QwenLikeModel-v7

without making the upstream Model Product an AI Product.

Authorized conclusion​

The same entity can be:

Product
+
Model Product
+
AI Primitive
+
contextual Business Product

while remaining outside the AI Product Family unless it independently satisfies the AIPS Product intelligence-proposition qualification.

Prohibited conclusions​

Productized model
-> AI Product automatically [INVALID]

Model Product used by AI Product
-> upstream Derived AI Product [INVALID]

M16-C02 - Counterexample​

QwenLikeModel-v7 is sold commercially
AND uses learned AI technology
therefore
QwenLikeModel-v7 is an AIPS AI Product

Invalid: business role, learned mechanism and Product status do not bypass AIPS AQ1-AQ5.

Invariants exercised​

M15-I03, M15-I04, M15-I05, M15-I07, M15-I12.


390. M16-P03 - Primary Symbolic Policy Reasoner​

Intent​

Demonstrate that enterprise AI is not synonymous with trained machine-learning models.

Participants​

  • AI Primitive classified SYMBOLIC_REASONER
  • Knowledge Resource
  • Rules
  • Reasoning and Classification Intelligence Capabilities
  • optional AI Product Version qualified as Primary

Canonical assertions​

PolicyReasonerPrimitive-v3
classifiedAsPrimitiveType SYMBOLIC_REASONER
contributesToIntelligenceCapability Reasoning
contributesToIntelligenceCapability Classification

PolicyReasonerProduct-v1.3
aips:dependsOn PolicyOntology-v8
aips:dependsOn Ruleset-v18

AI Execution E3
involvesAIPrimitive PolicyReasonerPrimitive-v3
exercisesIntelligenceCapability Reasoning

The reasoner may use deterministic logic and still satisfy AI Primitive semantics.

Authorized conclusion​

A trained model and retraining lifecycle are not universal prerequisites for enterprise AI or AI Product qualification.

Prohibited conclusions​

deterministic
-> not AI automatically [INVALID]

Ruleset-v18
-> SYMBOLIC_REASONER primitive automatically [INVALID]

Rule
== Symbolic Reasoner [INVALID]

M16-C03 - Counterexample​

PolicyReasonerProduct-v1.3 has learningMode NONE
therefore
it cannot be an AI Product

Invalid: mechanism-neutral AIPS semantics permit symbolic AI Products without ML retraining.

Invariants exercised​

M15-I07, M15-I12, M15-I17, M15-I36.


391. M16-P04 - Grounded / Retrieval-Augmented Generation Pattern​

Intent​

Represent a grounded generation pattern without minting RAG as an AI Primitive Type, Agent type or AI Product type.

Participants​

  • Retrieval activity
  • Knowledge Resource(s)
  • AI Context
  • resource playing Grounding Source Role
  • Prompt / Instruction
  • AI Primitive
  • AI Execution
  • Generation capability / GeneratedContent effect

Canonical assertions​

Retrieval R4
retrievesResource KnowledgeResource-K
suppliesToAIContext Context-C

KnowledgeResource-K
groundsAIContext Context-C

Prompt-P
containsInstruction Instruction-I

AI Execution E4
hasAIContext Context-C
involvesAIPrimitive GenerationPrimitive-G
exercisesIntelligenceCapability Generation
producesOrCausesIntelligenceEffect GeneratedContent

Authorized conclusion​

The pattern can express retrieval, grounding and generation while preserving independent identities and roles.

Prohibited conclusions​

Retrieval R4
-> INTELLIGENT_SEARCH automatically [INVALID]

retrievesResource K
-> K groundsAIContext automatically [INVALID]

RAG-like graph
-> Agent [INVALID]

RAG-like graph
-> AI Product automatically [INVALID]

Grounding Source
-> source is correct/authoritative automatically [INVALID]

M16-C04 - Counterexample​

Retrieval R4 retrievesResource K
therefore
K is trusted grounding evidence
AND the result is factually correct

Invalid: retrieval, grounding participation, source authority and correctness are separate assertions.

Invariants exercised​

M15-I02, M15-I05, M15-I10, M15-I11, M15-I19, M15-I27.


392. M16-P05 - Agent Uses an AI Product as a Tool​

Intent​

Demonstrate contextual Tool Role and Agent participation without ontology or authority leakage.

Participants​

  • Artificial Agent
  • AI Product playing Tool Role
  • AI Execution
  • Delegated Authority
  • Agent Autonomy Level / Control Mode under Agent Ontology
  • AIPS Product AR separately

Canonical assertions​

Agent-A
participatesInAIExecution E5
operatesUnderDelegatedAuthority DA5
usesTool FraudDetectionProduct-P

E5
usesTool FraudDetectionProduct-P

The tool resource remains an AI Product under AIPS while playing Tool Role in this context.

Authorized conclusion​

Agent identity, Tool Role, Product identity and authority remain independently governed.

Prohibited conclusions​

Agent usesTool AI Product
-> Agent is AI Product [INVALID]

Agent usesTool Product with ActionAuthorization
-> Agent inherits ActionAuthorization [INVALID]

Agent AutonomyLevel Autonomous
-> AIPS AR = AutonomousAction [INVALID]

M16-C05 - Counterexample​

Agent-A usesTool FraudDetectionProduct-P
therefore
Agent-A inherits P's Product governance profile,
risk tier and action authority

Invalid: relationship participation does not transfer Product governance or authority.

Invariants exercised​

M15-I01, M15-I02, M15-I12, M15-I17, M15-I18, M15-I19.


393. M16-P06 - Human-Reviewed Recommendation and Authorized Action​

Intent​

Demonstrate Recommendation -> Decision Formation -> Authorization -> Action without turning that sequence into a universal pipeline or equating human participation with approval.

Participants​

  • AI Execution
  • RecommendationEffect
  • Decision Formation
  • Human Agent
  • Authorization
  • DigitalAction or PhysicalAction occurrence where applicable

Canonical assertions​

AI Execution E6
producesOrCausesIntelligenceEffect RecommendationEffect

RecommendationOccurrence-R6
informsDecisionFormation DF6

HumanAgent-H6
participatesInDecisionFormation DF6

ActionOccurrence-A6
authorizedBy Authorization-6

Where applicable:

Authorization-6
withinDelegatedAuthority DA6

Authorized conclusion​

The AI recommendation can inform a human decision process, and a later action can be authorized, while Recommendation, Decision Formation, Authorization and Action remain separate.

Prohibited conclusions​

Human participatesInDecisionFormation
-> Human approved recommendation automatically [INVALID]

RecommendationEffect
-> Decision automatically [INVALID]

Decision effect
-> Authorization automatically [INVALID]

HumanInLoop
== ActWithApproval [INVALID]

M16-C06 - Counterexample​

AI produced RecommendationEffect
AND a Human Agent was present
therefore
the recommendation was reviewed, accepted and authorized

Invalid: participation does not prove review, acceptance or authorization.

Invariants exercised​

M15-I18, M15-I20, M15-I23, M15-I24, M15-I25.


394. M16-P07 - Autonomous Digital Action Under Bounded Authority​

Intent​

Represent autonomous digital action without interpreting AutonomousAction as unlimited permission or requiring a separate Decision effect in every occurrence.

Participants​

  • AI Product Version with AIPS AR AutonomousAction
  • DigitalAction Intelligence Effect
  • Agent
  • Delegated Authority
  • AI Execution
  • optional context-specific Authorization where policy requires it

Canonical assertions​

AI Product Version P7
AIPS AR = AutonomousAction
AIPS IE includes DigitalAction

Agent-A7
operatesUnderDelegatedAuthority DA7
participatesInAIExecution E7

E7
producesOrCausesIntelligenceEffect DigitalAction

If policy requires a concrete permission:

DigitalActionOccurrence-X7
authorizedBy AUTH7

AUTH7
withinDelegatedAuthority DA7

Authorized conclusion​

An autonomous action may occur under bounded Product/Agent authority without implying global authority.

Prohibited conclusions​

AutonomousAction
-> unlimited enterprise permission [INVALID]

DigitalAction
-> Decision effect necessarily occurred [INVALID]

DigitalAction
-> action succeeded [INVALID]

M16-C07 - Counterexample​

P7 has AR = AutonomousAction
therefore
P7 may invoke any Tool and perform any digital action
in any domain without additional scope constraints

Invalid: autonomy classification is bounded by Product Promise, policy, delegated authority, governance and applicable authorization conditions.

Invariants exercised​

M15-I18, M15-I19, M15-I23, M15-I24, M15-I25, M15-I29.


395. M16-P08 - Closed-Loop Physical AI​

Intent​

Demonstrate Physical AI while keeping Embodiment, sensing, Perception, Control, Actuation, PhysicalAction, Agent identity and authorization distinct.

Participants​

  • AI Product Version
  • AIPS EM ClosedLoopAutonomous
  • AIPS AR AutonomousAction
  • AIPS IE PhysicalAction
  • Environment
  • Control Loop
  • Agent
  • AI Execution
  • sensing / acting relationships

Canonical assertions​

NavigationProduct-P8
AIPS EM = ClosedLoopAutonomous
AIPS AR = AutonomousAction
AIPS IE includes PhysicalAction

RoboticAgent-A8
participatesInAIExecution E8

E8
occursInEnvironment Warehouse-8
sensesEnvironment Warehouse-8
participatesInControlLoop Loop-8
actsUponEnvironment Warehouse-8
producesOrCausesIntelligenceEffect PhysicalAction

AIPS Product governance separately computes the applicable risk floor and action/safety obligations.

Authorized conclusion​

A physical closed-loop AI system can be represented without collapsing environmental interaction into Product Embodiment or action authority.

Prohibited conclusions​

RoboticAgent
-> AIPS EM = EmbodiedActuating automatically [INVALID]

sensesEnvironment
-> Perception capability automatically [INVALID]

actsUponEnvironment
-> authorized PhysicalAction automatically [INVALID]

Control Loop
-> ClosedLoopAutonomous automatically [INVALID]

M16-C08 - Counterexample​

E8 actsUponEnvironment Warehouse-8
therefore
NavigationProduct-P8 is ClosedLoopAutonomous
AND every physical action is authorized

Invalid: occurrence interaction does not replace AIPS Product-version Embodiment classification or authorization semantics.

Invariants exercised​

M15-I08, M15-I18, M15-I21, M15-I22, M15-I25, M15-I29.


396. M16-P09 - Derived / Composite Integrated Transaction Risk AI Product​

Intent​

Demonstrate AIPS Derived APT and Composite derivation topology without confusing them with generic GEAIRM Composition.

Participants​

  • Fraud Detection AI Product
  • Sanctions Intelligence AI Product
  • Behavioural Risk AI Product
  • Integrated Transaction Risk AI Product
  • AIPS materiallyDependsOn
  • APT Derived
  • Derivation Topology Composite

Canonical assertions​

IntegratedRisk-D
aips:materiallyDependsOn Fraud-A
aips:materiallyDependsOn Sanctions-B
aips:materiallyDependsOn Behaviour-C

IntegratedRisk-D
AIPS Product qualification = CURRENT
APT = Derived
Derivation Topology = Composite

Generic GEAIRM Composition may additionally model an operational arrangement, but that assertion is independent.

Authorized conclusion​

The Product is Derived because its distinct governed intelligence proposition materially depends on upstream AI Products; Composite describes the multi-upstream derivation topology.

Prohibited conclusions​

Composite
-> third APT beside Primary/Derived [INVALID]

three upstream APIs
-> Derived automatically [INVALID]

GEAIRM Composition
-> AIPS Composite automatically [INVALID]

M16-C09 - Counterexample​

Service S calls three AI Products
therefore
S is a Composite Derived AI Product

Invalid: runtime calls do not establish Productness, AI Product qualification, material intelligence provenance or a distinct governed intelligence proposition.

Invariants exercised​

M15-I03, M15-I04, M15-I12, M15-I15, M15-I16.


397. M16-P10 - Runtime Orchestration Without New AI Product Identity​

Intent​

Demonstrate that orchestration of several governed AI Products can exist as a runtime/semantic arrangement without creating another Product.

Participants​

  • Orchestration
  • Orchestrating Agent or workflow
  • multiple AI Products
  • Tool/resource participants

Canonical assertions​

Agent-O10
participatesInOrchestration O10

AIProduct-P1
participatesInOrchestration O10

AIProduct-P2
participatesInOrchestration O10

Tool-T10
participatesInOrchestration O10

Authorized conclusion​

The arrangement can coordinate Products and Tools while every participant retains its own identity.

Prohibited conclusions​

Orchestration O10
-> AI Product automatically [INVALID]

O10 has multiple Products
-> AIPS Composite automatically [INVALID]

O10 coordinates enterprise participants
-> AIPS EnterpriseOrchestrated automatically [INVALID]

M16-C10 - Counterexample​

O10 orchestrates P1 and P2
therefore
O10 is a Derived Composite AI Product

Invalid absent BPS Productness and AIPS qualification of a distinct governed intelligence proposition.

Invariants exercised​

M15-I02, M15-I03, M15-I04, M15-I15, M15-I16.


398. M16-P11 - Multi-Agent Composition Without Collective-Agent Inference​

Intent​

Demonstrate cooperative multi-Agent behaviour without assuming one Collective Agent, Product or shared authority.

Participants​

  • Human, Artificial or Hybrid Agents
  • GEAIRM Composition
  • optional Orchestration
  • individual Agent authority/autonomy semantics

Canonical assertions​

Agent-A11
participatesInComposition C11

Agent-B11
participatesInComposition C11

Agent-C11
participatesInComposition C11

Optionally:

Agent-A11
participatesInOrchestration O11

Agent-B11
participatesInOrchestration O11

Authorized conclusion​

Multiple independently identified Agents can cooperate while retaining distinct identity, autonomy, authority, memory and lifecycle.

Prohibited conclusions​

three Agents
-> CollectiveAgent automatically [INVALID]

multi-Agent Composition
-> Compound Product automatically [INVALID]

Agent A authority
-> Agent B authority [INVALID]

M16-C11 - Counterexample​

A11, B11 and C11 participateInComposition C11
therefore
C11 is one Autonomous Cognitive Collective Agent
with the union of all Agent permissions

Invalid: Collective Agent identity and authority require independent semantic qualification.

Invariants exercised​

M15-I02, M15-I05, M15-I17, M15-I18, M15-I19.


399. M16-P12 - AI-Enabled Streaming Service Remains a Service Product​

Intent​

Stress-test BPS PIN and composition non-inheritance.

Participants​

  • Streaming Service Product
  • Recommendation AI Product
  • Data Products
  • content Products
  • platform capabilities

Canonical assertions​

StreamingService-S12
BPS Product Family = Service Product

StreamingService-S12
aips:dependsOn RecommendationAI-P12

The recommendation capability may be essential to competitive value, but the intrinsic Product Family still follows BPS PIN: ongoing managed service/access is the value-bearing essence.

Authorized conclusion​

A Product can be heavily AI-enabled without becoming an AI Product.

Prohibited conclusions​

Service dependsOn AI Product
-> Service becomes AI Product [INVALID]

AI is commercially important
-> Product Family = AI Product [INVALID]

M16-C12 - Counterexample​

RecommendationAI-P12 is essential to user engagement
therefore
StreamingService-S12 inherits AI Product Family

Invalid: constitution and dependencies inform the Product but do not automatically determine Product Family. BPS explicitly preserves Product intrinsic nature independently of component technology and composition.

Invariants exercised​

M15-I03, M15-I04, M15-I12, M15-I15.


400. M16-P13 - Evaluation and Assurance Without Risk Collapse​

Intent​

Demonstrate Evaluation, Evidence and Assurance while preserving AIPS Effective Risk Tier as an independent governance computation.

Participants​

  • Evaluation
  • Evaluation Criterion
  • Evaluation Result
  • AIPS AI Evaluation Report
  • Evidence Role
  • Assurance Assessment
  • Assurance Finding
  • AIPS Effective Risk Tier

Canonical assertions​

Evaluation-E13
evaluatesTarget FraudProduct-P13
usesEvaluationCriterion CalibrationCriterion
producesEvaluationResult Result-R13

Result-R13
recordedInEvaluationReport Report-13

AssuranceAssessment-AA13
assessesAssuranceOf FraudProduct-P13
usesEvidence Report-13
producesAssuranceFinding Finding-F13

FraudProduct-P13
AIPS Effective Risk Tier = R3

Authorized conclusion​

A strong assurance finding and R3 risk posture can coexist because assurance evidential sufficiency and risk classification answer different questions.

Prohibited conclusions​

Assurance PASS
-> Effective Risk Tier = R0/R1 [INVALID]

AIEvalReport exists
-> assurance sufficient [INVALID]

Evidence present
-> governance obligation satisfied [INVALID]

M16-C13 - Counterexample​

All evaluation criteria passed
therefore
Product risk tier must be reduced to R0

Invalid: AIPS Effective Risk Tier is computed from authoritative Product-version risk drivers, not directly from an M13 Evaluation/Assurance result.

Invariants exercised​

M15-I26, M15-I27, M15-I28, M15-I29, M15-I30.


401. M16-P14 - Dependency Currentness Change Without Historical Mutation​

Intent​

Demonstrate pinned version semantics, Currentness Assessment and governed reassessment.

Participants​

  • Downstream Product Version D7
  • Upstream Product Version U3
  • newer upstream U4
  • Dependency Path / binding
  • Currentness Assessment
  • qualification/dependency currentness

Canonical assertions​

At time T1:

Downstream-D7
aips:dependsOn Upstream-U3

CurrentnessAssessment-14A
assessesCurrentnessOf Downstream-D7
hasCurrentnessConcern dependency-currentness
hasCurrentnessStatus CURRENT

At time T2, U4 is published.

No semantic mutation of D7 follows automatically.

If a policy or upstream-state change materially invalidates the U3 binding, a later assessment may state:

CurrentnessAssessment-14B
assessesCurrentnessOf Downstream-D7
hasCurrentnessConcern dependency-currentness
hasCurrentnessStatus REVALIDATION_REQUIRED

Authorized conclusion​

Historical binding to U3 remains true. Currentness can change later without rewriting that historical dependency.

Prohibited conclusions​

U4 published
-> D7 now dependsOn U4 automatically [INVALID]

newer upstream exists
-> D7 STALE automatically [INVALID]

currentness changes
-> rewrite historical D7 graph [INVALID]

M16-C14 - Counterexample​

U4 is newer than U3
therefore
replace every historical D7 -> U3 dependency
with D7 -> U4

Invalid: this destroys reproducibility and violates M14 historical non-mutation.

Invariants exercised​

M15-I12, M15-I13, M15-I31, M15-I32, M15-I33, M15-I35.


402. M16-P15 - AI Primitive Removal and Product-Family Reclassification​

Intent​

Demonstrate that Product identity can continue while Product-version AI qualification changes, without retroactively rewriting the historical AI Product version.

Participants​

  • continuing Product Identity A
  • A v1
  • A v2
  • Qwen-like Model Product / AI Primitive in v1
  • deterministic if/else implementation in v2
  • AIPS Qualification Status
  • Product Family / APT
  • downstream versions pinned to v1 or adopting v2

Canonical assertions​

Historical v1:

Product Identity A
aips:hasVersion A-v1

A-v1
has qualifying AI Primitive path to ModelProduct-M1
AI Product qualification = CURRENT
APT = Primary

Later v2:

Product Identity A
aips:hasVersion A-v2

A-v2
replaces the only qualifying AI Primitive
with ordinary deterministic if/else logic
qualification for AI Product Family = INVALID
no current APT

Historical interpretation:

A-v1 remains historically an AI Product.
A-v2 is classified from its own constitution.

Downstream:

Downstream-X7 pinned to A-v1
-> historical/current interpretation remains against A-v1
unless its own currentness changes for another governed reason

Downstream-X8 adopts A-v2
-> must recompute its applicable qualification/dependency semantics

Authorized conclusion​

Reclassification is version-specific. Continuing Product identity does not force AI Product Family continuity.

Prohibited conclusions​

A-v2 is no longer AI
-> A-v1 was never AI [INVALID]

same Product Identity
-> v2 must keep APT = Primary [INVALID]

A-v2 invalid qualification
-> every downstream version pinned to A-v1 invalid [INVALID]

M16-C15 - Counterexample​

A-v2 removes the final AI Primitive
therefore
rewrite A-v1 Product Family from AI Product
to conventional software Product

Invalid: historical classification remains true for A-v1. AIPS lifecycle examples explicitly preserve v1 as historical AI Product evidence while v2 may become non-AI. The governing BPS ontology also separates intrinsic nature from lifecycle/context changes rather than allowing later context to redefine prior intrinsic identity.

Invariants exercised​

M15-I04, M15-I31, M15-I32, M15-I33, M15-I34, M15-I35.


403. Cross-Pattern Counterexample Catalogue​

The fifteen counterexamples can be summarized as prohibited inference families:

CounterexampleInvalid shortcut
M16-C01Model Product dependency -> Derived APT
M16-C02Productized learned model -> AI Product
M16-C03no ML/retraining -> not AI
M16-C04retrieval -> trusted grounding/correctness
M16-C05Tool use -> inherited Product authority/governance
M16-C06human presence -> reviewed/accepted/authorized
M16-C07AutonomousAction -> unlimited authority
M16-C08environment action occurrence -> Product Embodiment/authorization
M16-C09calls multiple AI Products -> Composite Derived AI Product
M16-C10Orchestration -> new AI Product
M16-C11multi-Agent Composition -> Collective Agent with union authority
M16-C12AI dependency -> Product Family inheritance
M16-C13successful Evaluation/Assurance -> low Risk Tier
M16-C14newer upstream -> rewrite historical dependency binding
M16-C15later reclassification -> rewrite historical Product Family

These counterexamples are not exhaustive. They represent recurring failure modes that the M15 invariants are intended to prevent.


404. Cross-Pattern Semantic Coverage​

M16 confirms that the same reference model can represent:

Learned model AI
Symbolic AI
Knowledge/reasoning AI
Grounded generation / retrieval
Agentic AI
Tool-using Agents
Human-reviewed decision support
Autonomous digital action
Physical / embodied AI
Derived AI Products
Composite derivation
Multi-Agent systems
AI-enabled non-AI Products
Evaluation / Assurance
Currentness / historical reclassification

without requiring:

one universal ML model assumption
one universal Agent assumption
one mandatory Product boundary
one mandatory recommendation->decision->action pipeline
one universal risk inference
one universal architecture topology

This is a central 0.1 design outcome.

M16-R05 - Mechanism breadth: Canonical examples MUST preserve GEAIRM mechanism neutrality and MUST NOT imply that learned models are required for enterprise AI.

M16-R06 - Product optionality: Canonical examples MUST preserve the fact that enterprise-AI participants may exist without being Products.

M16-R07 - Agent optionality: Canonical examples MUST NOT imply that every AI Execution requires an Agent or that every Agent is AI.


405. Pattern-to-Invariant Coverage​

The following table confirms broad invariant coverage.

Invariant familyPatterns exercising it
Authority preservation / scopeP01, P02, P05, P09, P13, P15
Product non-inferenceP02, P04, P05, P09, P10, P11, P12
Classification orthogonalityP01, P03, P07, P08, P13
Occurrence/Product separationP01, P06, P08
Context/Role non-intrinsicalityP04, P05, P11
Dependency non-inheritanceP01, P05, P09, P12, P14
Composition non-inheritanceP09, P10, P11, P12
Agent/AI separationP05, P11
Autonomy/Authority separationP05, P06, P07, P08, P11
Decision/Authority separationP06, P07
Evaluation/Evidence/Assurance separationP13
Assurance/Risk orthogonalityP13
Historical non-mutationP14, P15
Currentness contextualityP14, P15
Provenance/dependency disciplineP14, P15
Architecture boundaryall patterns

M16 is not intended to prove formal completeness of the invariant set. It demonstrates representative semantic coverage before M17 consolidation and the later A3 executable invariant catalogue.


406. Pattern Conformance Interpretation​

A model instance is consistent with an M16 pattern when its relevant assertions preserve the normative concepts, authorities, relationships and invariants represented by that pattern.

Pattern consistency does not mean:

  • complete GEAIRM conformance;
  • AIPS Product qualification;
  • AIPS deployment conformance;
  • regulatory approval;
  • production readiness;
  • architecture approval.

Likewise, an M16 counterexample being rejected does not prove the remaining graph is fully conformant.

M16 therefore distinguishes:

pattern consistency
!=
full semantic conformance
!=
AIPS Product qualification
!=
GEAIRA architecture conformance

M16-R08 - Pattern/conformance distinction: M16 pattern consistency MUST NOT be represented as complete GEAIRM or AIPS conformance.

M16-R09 - Counterexample rejection restraint: Rejection of an M16 counterexample MUST NOT be represented as proof that an entire model instance is valid.


407. Architecture Boundary in Examples​

M16 examples are semantic patterns.

They deliberately do not prescribe:

Model Registry
Agent Runtime
Prompt Store
Vector Database
Context Service
Tool Registry
Policy Engine
Authorization Service
Decision Engine
Evidence Store
Risk Engine
Control Plane
Orchestration Platform
Digital Twin
Robotics Middleware
Lineage Store
Version Registry

Any of those may be legitimate architectural choices in GEAIRA or an enterprise implementation, but they cannot be inferred solely from the GEAIRM semantic relationships.

For example:

usesTool
!= requires Tool Registry

retrievesResource
!= requires vector database

groundsAIContext
!= requires RAG framework

authorizedBy
!= requires one particular authorization service

usesEvidence
!= requires centralized evidence store

supersedes
!= requires source-control implementation

M16-R10 - Architecture non-inference: M16 patterns MUST NOT be used to infer mandatory GEAIRA capabilities, ABBs, registries, platforms or technologies.


408. Canonical Pattern Invariants​

M16 adds no new semantic invariant family beyond M15. It instead establishes pattern-specific application rules:

RuleRequirement
M16-R11Learned-model examples MUST preserve Model Product / AI Product distinction.
M16-R12Symbolic examples MUST remain valid without mandatory learned-model/retraining assumptions.
M16-R13Grounded-generation examples MUST keep Retrieval, Grounding, Context and correctness distinct.
M16-R14Agent/Tool examples MUST preserve Tool Role and authority separation.
M16-R15Human-review examples MUST NOT infer approval or acceptance from Human participation alone.
M16-R16Autonomous-action examples MUST preserve bounded authority and MUST NOT imply unrestricted permission.
M16-R17Physical-AI examples MUST preserve Sensing/Perception, Control/Actuation/PhysicalAction and Embodiment distinctions.
M16-R18Derived/Composite Product examples MUST preserve AIPS APT and Derivation Topology authority.
M16-R19Runtime orchestration MUST NOT create AI Product identity absent BPS/AIPS qualification.
M16-R20Multi-Agent examples MUST NOT infer Collective Agent, Compound Product or shared authority solely from participation.
M16-R21AI-enabled Product examples MUST preserve BPS PIN and composition non-inheritance.
M16-R22Evaluation/Assurance examples MUST preserve Evidence non-proof and Assurance/Risk orthogonality.
M16-R23Currentness examples MUST preserve bound-version historical truth and contextual currentness.
M16-R24Reclassification examples MUST remain version-specific and MUST NOT retroactively relabel historical versions.
M16-R25Product-Version declarations and occurrence assertions MUST remain distinct in every pattern.
M16-R26Contextual Roles in examples MUST NOT be promoted to intrinsic types.
M16-R27Counterexamples MUST preserve the distinction between semantic invalidity and formal OWL disjointness.
M16-R28Example labels MUST NOT be treated as identity or equivalence evidence.
M16-R29Dependency reachability in examples MUST preserve M11 Dependency Path semantics rather than universal transitivity.
M16-R30Governance/risk semantics MUST NOT propagate through example dependency/composition graphs by inheritance.
M16-R31Runtime observations in examples MUST NOT mutate Product ontology directly.
M16-R32Historical examples MUST preserve superseded/retired evidence where required by the applicable authority.
M16-R33Example use of external concepts/properties MUST preserve the source authority's scope and cardinality.
M16-R34M16 introduces no new GEAIRM relationship type; any later new relationship requires explicit normative amendment.
M16-R35M16 introduces no new Product, Agent, risk, lifecycle or architecture taxonomy.
M16-R36M16 pattern identifiers are documentation identifiers and MUST NOT be treated as executable namespace identifiers.
M16-R37Machine-readable test cases corresponding to these patterns remain deferred to A9-A14.
M16-R38M16 remains technology-neutral and implementation-independent.

Together with M16-R01 through M16-R10, M16 establishes 38 normative rules.


409. M16 Exit Criteria​

GEAIRM-M16 is complete when:

  • learned-model AI is represented without collapsing Model Product and AI Product;
  • a Productized model boundary case is explicitly tested;
  • symbolic AI is represented without mandatory ML/retraining assumptions;
  • grounded generation is represented without minting RAG as a Product/Agent/primitive type;
  • Retrieval and Grounding remain distinct;
  • Agent/Tool participation preserves Product and authority boundaries;
  • human-reviewed recommendation/decision semantics do not infer approval from presence;
  • autonomous digital action preserves bounded authority;
  • Physical AI preserves Environment, Sensing, Control Loop, Embodiment and PhysicalAction distinctions;
  • Derived/Composite AIPS Product semantics are represented without confusing generic Composition;
  • runtime Orchestration does not create a new Product by graph shape;
  • multi-Agent Composition does not create Collective Agent or shared authority automatically;
  • an AI-enabled Service Product preserves its BPS Product Family;
  • Evaluation, Evidence, Assurance and Risk remain independent;
  • dependency currentness change preserves bound historical versions;
  • primitive removal/reclassification preserves historical AI Product truth;
  • all examples use M15 canonical relationships or externally authoritative bridged properties;
  • all counterexamples violate already established semantics rather than inventing new semantics;
  • the M15 invariant catalogue receives broad scenario coverage;
  • pattern consistency is distinguished from full conformance;
  • no architecture components are inferred from semantic relationships;
  • no new domain concept or relationship is introduced by M16;
  • M01-M15 authority, Product boundary, conservative inference and historical-preservation principles remain intact.

M16 conclusion: these criteria are satisfied by this slice. GEAIRM 0.1 now demonstrates that its semantic model can represent learned, symbolic, grounded, agentic, autonomous, Physical, Derived, multi-Agent, evaluated and evolving enterprise AI while rejecting recurrent ontology, authority, Product and historical-inference errors.


410. Forward Work​

The next slice is:

GEAIRM-M17 - Reference Views, Alignment & 0.1 Consolidation

M17 will consolidate the full GEAIRM 0.1 model into canonical reference views, map the final domain structure, reconcile terminology and authority boundaries across BPS, AIPS, the Agent Ontology, GEDRM/GEARM and PROV, confirm the M01-M16 exit criteria, and freeze the human-readable 0.1 semantic baseline before executable formalization begins with A1.


411. GEAIRM-M17 - Reference Views, Alignment & 0.1 Consolidation​

411.1 Purpose of this slice​

M01-M16 established the semantic jurisdiction, metamodel, core Artificial Intelligence semantics, mechanism/capability model, execution/effect semantics, context/grounding model, Agent/authority model, physical-AI semantics, decision/action semantics, dependency/composition model, Product boundary, evaluation/assurance/governance model, temporal/provenance model, canonical relationship/invariant baseline and representative patterns/counterexamples.

M17 performs the final human-readable consolidation for GEAIRM 0.1.

M17 answers seven questions:

  1. What is the final GEAIRM 0.1 semantic-domain structure?
  2. What canonical reference views should be used to understand the model without turning views into ontology classes?
  3. How does GEAIRM align with BPS, AIPS, the KivaNura Agent Ontology, GEDRM/GEDRA, GEARM/GEARA and W3C PROV?
  4. Which semantic concerns are owned by GEAIRM and which remain externally authoritative?
  5. What cross-cutting invariants apply to the entire model?
  6. What exactly is frozen at the end of M17?
  7. What is handed to the executable A1-A15 formalization programme?

M17 introduces no new enterprise-AI domain concept and no new GEAIRM relationship type. Its job is to organize, reconcile and freeze the human-readable semantic baseline already established.

M17-R01 - Consolidation-only rule: M17 MUST consolidate M01-M16 and MUST NOT silently create new semantic meaning that bypasses the authority, relationship or invariant disciplines already established.


412. Final GEAIRM 0.1 Semantic Domain Map​

The provisional domain hypothesis from the programme is now replaced by the following final organizational structure.

These semantic domains are reference-model views, not ontology superclasses, Product types, architecture planes or deployment layers.

DomainFinal labelPrimary concerns
D1Core AI SemanticsArtificial Intelligence, Intelligence Capability, qualifying AI semantics, AI/automation boundary
D2Mechanism, Execution & EffectAI Primitive, AI Primitive Type, AI Execution, Execution Mode, Intelligence Effect
D3Context, Knowledge, Grounding & MemoryAI Context, contextual resources, Knowledge Resources, Rules, Prompt/Instruction, Retrieval, Grounding, Memory
D4Agency, Authority, Decision & ActionAgent participation, Tool Role, autonomy/control distinctions, Delegated Authority, Decision Formation, Authorization, Recommendation/Plan/Decision/Action boundaries
D5Environment, Embodiment & ControlEnvironment, Sensing, physical-world interaction, Control Loop, AIPS Embodiment alignment, Actuation/PhysicalAction separation
D6Dependency, Composition, Product & Topologydependency paths, material dependency, Composition, Federation, Orchestration, BPS/AIPS Product boundary, APT, Derivation Topology, Topology Scope
D7Evaluation, Assurance, Risk & GovernanceEvaluation, Evidence Role, Assurance Assessment/Finding, AIPS Product Assurance, Fitness for Purpose, R0-R4, Risk Driver Assessment, Governance Obligation/Profile
D8Identity, Lifecycle, Currentness, Provenance & Changecontinuing identity, versioned resources, lifecycle states, currentness, supersession, PROV/AIPS provenance, lineage, semantic change impact

412.1 Domain-map interpretation​

The domains are orthogonal organizing lenses over one graph.

A participant may appear in several domains without acquiring multiple intrinsic types. For example:

AI Product Version
appears in D2 through declared capability/mode/effect
appears in D6 through Product/dependency semantics
appears in D7 through risk/governance
appears in D8 through version/currentness/lifecycle

This is one entity viewed through multiple semantic concerns.

Likewise:

Agent
appears in D4 as an Agent
may participate in D2 AI Execution
may interact with D5 Environment
may participate in D6 Composition
may be evaluated in D7
may have version/lifecycle semantics in D8

No domain membership is itself an ontology classification.

M17-R02 - Domain-view restraint: D1-D8 MUST be treated as organizational reference-model views and MUST NOT be encoded as intrinsic semantic types merely because an element appears within a domain.

M17-R03 - Cross-domain identity preservation: An element appearing in multiple GEAIRM domains MUST retain one governed identity under its semantic authority unless an independent classification rule establishes otherwise.


413. Cross-Cutting Semantic Foundations​

Four concerns cut across all eight domains.

413.1 Semantic Authority​

Every canonical element must have an identifiable semantic authority or an explicit GEAIRM ownership disposition.

413.2 Product Boundary​

Productness and Product Family remain independent of AI participation. BPS remains authoritative for generic Product ontology; AIPS remains authoritative for AI Product qualification.

413.3 Conservative Inference​

Relationship assertions do not silently create identity, authority, Product status, risk, lifecycle or architecture semantics.

413.4 Historical Truth​

Later change, currentness degradation, supersession or reclassification does not rewrite historical version truth.

The cross-cutting foundation is:

Semantic Authority
+
Product Boundary
+
Conservative Inference
+
Historical Preservation
|
v
all D1-D8 interpretations

M17-R04 - Cross-cutting foundation: Every GEAIRM domain and reference view MUST preserve authority, Product-boundary, conservative-inference and historical-preservation rules.


414. Canonical Reference View V1 - Authority Landscape​

The first canonical view answers:

Who owns which meaning?

BPS / PIN
Product
Product Intrinsic Nature
Product Family / Kind
contextual Product roles
Compound Product
|
| specializes into
v
AIPS
AI Product qualification
AI Product Version / APT
AI Primitive / Primitive Type
Intelligence Capability
Execution Mode / Effect
Agency-Autonomy / Embodiment
Topology Scope
Risk / Governance / Product lifecycle
|
| integrated by
v
GEAIRM
generic enterprise-AI semantic graph
+ GEAIRM-owned gap concepts/relationships

KivaNura Agent Ontology
Agent identity / taxonomy
Agent autonomy / control / tool-use pattern
|
+------------------> GEAIRM integration

GEDRM / GEDRA
Data semantics / Data architecture
|
+------------------> GEAIRM cross-domain reference

GEARM / GEARA
Information / Analytics semantics and architecture
|
+------------------> GEAIRM cross-domain reference

W3C PROV
generic provenance semantics
|
+------------------> GEAIRM/AIPS provenance use

GEAIRM is therefore an integrating semantic reference model, not an authority takeover layer.

M17-R05 - Authority-landscape preservation: The V1 authority landscape MUST remain authoritative for GEAIRM 0.1 formalization unless a later normative revision explicitly changes an ownership boundary.


415. Canonical Reference View V2 - Core AI Semantic Chain​

The core AI view answers:

What constitutes the minimum semantic nucleus of enterprise AI?

Artificial Intelligence
|
| characterized through
v
Intelligence Capability
^
|
| contributesToIntelligenceCapability
|
AI Primitive
|
| classifiedAsPrimitiveType
v
AI Primitive Type

The runtime occurrence view adds:

AI Execution
|
+-- involvesAIPrimitive ------------> AI Primitive
|
+-- exercisesIntelligenceCapability -> Intelligence Capability
|
+-- operatesInExecutionMode --------> Execution Mode
|
+-- producesOrCausesIntelligenceEffect
-> Intelligence Effect

The interpretation remains:

mechanism  != capability
capability != mode
mode != effect
effect != Product

M17-R06 - Core-axis preservation: The V2 mechanism/capability/execution/effect axes MUST remain orthogonal and MUST NOT be collapsed during executable formalization.


416. Canonical Reference View V3 - Context, Grounding and Knowledge​

The context view answers:

What information, knowledge and retained state are made relevant to an AI Execution, and how?

                      AI Execution
|
hasAIContext
|
v
AI Context
/ | \
/ | \
Data / Information Knowledge / Rule Prompt / Memory
cross-domain refs resources resources
\ | /
\ | /
+---- contextual resources +
|
Grounding Source Role
^
|
groundsAIContext

Retrieval
+-- retrievesResource ------> resource
+-- suppliesToAIContext ----> AI Context

The view preserves:

Context != Grounding
Retrieval != Grounding
Retrieval != Intelligent Search
Memory != learning
Prompt != Instruction
Instruction != Authorization
Knowledge Resource != Knowledge Reasoner

Data and Information semantics remain externally governed rather than redefined.

M17-R07 - Context-boundary preservation: The V3 context view MUST preserve the distinction among resource inclusion, retrieval, grounding, memory, instruction and external Data/Information authority.


417. Canonical Reference View V4 - Agent, Authority, Decision and Action​

The agency view answers:

Who participates, how independently do they operate, under what authority, and how are decisions/actions related?

Agent
|
+-- participatesInAIExecution --------> AI Execution
|
+-- operatesUnderDelegatedAuthority --> Delegated Authority
|
+-- usesTool --------------------------> resource playing Tool Role

Agent Ontology:
Agent Autonomy Level
Agent Control Mode
Tool Use Pattern

AIPS:
Agency / Autonomy
Inform | Recommend | Assist | ActWithApproval | AutonomousAction

Decision/action view:

Recommendation / Plan / Evidence
|
v
Decision Formation
/ \
/ \
Agent / AI other inputs
Execution
|
v
Decision effect
does not itself
authorize
|
v
Authorization
|
authorizedBy
|
v
Action occurrence

The canonical separations are:

Agent autonomy
!= AIPS Agency/Autonomy
!= Control Mode
!= Delegated Authority
!= Authorization

Recommendation != Decision
Decision effect != Decision authority
Decision effect != Authorization
Authorization != Action
Action != successful outcome

M17-R08 - Agency/authority preservation: The V4 view MUST preserve Agent identity, autonomy, control, Delegated Authority, Authorization and AIPS Product Agency/Autonomy as distinct semantics.

M17-R09 - Decision/action preservation: Recommendation, Plan, Decision Formation, Decision effect, Authorization and Action MUST remain independently representable and MUST NOT be forced into one universal pipeline.


418. Canonical Reference View V5 - Environment and Physical AI​

The physical-AI view answers:

How does an AI occurrence sense and affect an Environment without collapsing physical interaction into Product Embodiment or authority?

Environment
^
|
sensesEnvironment
|
Agent / AI Execution
|
+-- participatesInControlLoop --> Control Loop
|
+-- actsUponEnvironment -------> Environment
|
+-- may produce PhysicalAction

Aligned but distinct Product classification:

AIPS Embodiment
NonEmbodied
PhysicalWorldPerceptive
EmbodiedActuating
ClosedLoopAutonomous

The view preserves:

Sensing != Perception
Control capability != Control Loop
Control capability != Actuation
Actuation != PhysicalAction
RoboticAgent != EmbodiedActuating
Control Loop != ClosedLoopAutonomous
physical capability != permission

M17-R10 - Physical-AI distinction preservation: V5 MUST preserve Environment interaction, Embodiment, Agent identity, Control, Actuation, PhysicalAction and authority as separate semantic concerns.


419. Canonical Reference View V6 - Dependency, Product and Coordination​

The structural/Product view answers:

What does a participant rely upon, how is it composed/coordinated, and when does that become a governed Product proposition?

                 DEPENDENCY

A -- dependsOn ----------------------> B
A -- materiallyDependsOn -----------> C

Dependency Path:
A -> B -> C -> Primitive P

COMPOSITION

P1 ----\
P2 -----+--> Composition C
P3 ----/

FEDERATION

independently governed participants
|
+--> Federation F

ORCHESTRATION

participants
|
+--> Orchestration O

Product bridge:

BPS Productness / PIN
|
v
AIPS AI Product qualification
AQ1 Productness
AQ2 Intelligence primacy
AQ3 Distinct intelligence proposition
AQ4 Qualifying AI Primitive path
AQ5 Governed Product boundary
|
v
AI Product Version
|
+--> APT Primary | Derived
|
if Derived:
SingleUpstream | Composite

AIPS Topology Scope remains:

Standalone
DomainCoupled
CrossDomainFederated
EnterpriseOrchestrated

and does not become deployment topology.

M17-R11 - Structural/Product boundary: Dependency, Composition, Federation and Orchestration MUST NOT create Product or AI Product identity without BPS/AIPS qualification.

M17-R12 - Product-classification authority: APT, Derivation Topology and Topology Scope MUST remain AIPS-governed Product classifications and MUST NOT be generalized into generic GEAIRM topology types.


420. Canonical Reference View V7 - Evaluation, Assurance, Risk and Governance​

The trust/governance view answers:

What was assessed, what evidence supports it, what assurance can be concluded, and what AIPS risk/governance regime applies?

Evaluation
|
+-- evaluatesTarget -----------> target
+-- usesEvaluationCriterion ---> criterion
+-- usesEvidence --------------> evidence-role resource
+-- producesEvaluationResult --> Evaluation Result
|
v
AIPS AIEvalReport

Assurance Assessment
|
+-- assessesAssuranceOf -------> target/claim
+-- usesEvidence --------------> evidence-role resource
+-- producesAssuranceFinding --> Assurance Finding

AIPS Product trust/governance remains external authority:

AI Product Version
|
+-- Product Assurance
+-- Fitness for Purpose
+-- Effective Risk Tier R0-R4
+-- Risk Driver Assessment
+-- Governance Obligation Profile
|
+-- Governance Obligations
|
v
AIPDS Governance Control Binding

Canonical distinction:

Evaluation != Assurance
Evidence != proof
Assurance != Risk
Risk Tier != Fitness for Purpose
Governance Obligation != Governance Control

M17-R13 - Trust-axis preservation: Evaluation, Evidence, Assurance, Fitness, Risk and Governance MUST remain separate axes connected only through explicitly governed relationships/rules.


421. Canonical Reference View V8 - Identity, Currentness and Evolution​

The temporal view answers:

How does meaning change over time without rewriting history?

Continuing identity
|
+---- governed version v1
|
+---- governed version v2
|
+---- governed version v3

later version
-- supersedes --> earlier version

Currentness Assessment
-- assessesCurrentnessOf --> target
-- hasCurrentnessConcern -> concern
-- hasCurrentnessStatus --> authority-specific status

Semantic Change Event
-- affectsSemanticTarget --> participant/version

Provenance and lineage remain:

W3C PROV / AIPS provenance
explains origin and derivation

Dependency graph
explains reliance

Version history
explains revision/succession

Lineage
traces these explicit relationships
without collapsing their meanings

The core invariant remains:

Historical semantic truth is preserved; currentness is reassessed.

M17-R14 - Temporal preservation: V8 MUST preserve identity/version separation, concern-specific currentness, historical non-mutation, provenance/dependency separation and scoped supersession.


422. Final Authority and Disposition Matrix​

The following matrix is the consolidated M17 authority boundary.

Semantic concernAuthorityGEAIRM 0.1 disposition
Generic Productness / PINBPSREFERENCED
Product Family / Product KindBPSREFERENCED
Contextual Product rolesBPSREFERENCED
Compound ProductBPSREFERENCED
AI Product qualification / AQEAIPSREUSED / REFERENCED
AI Product Type / Derivation TopologyAIPSREUSED
AI Primitive / AI Primitive TypeAIPSREUSED
Intelligence CapabilityAIPSREUSED
Execution ModeAIPSREUSED
Intelligence EffectAIPSREUSED
Agency / AutonomyAIPSREUSED
EmbodimentAIPSREUSED
Topology ScopeAIPSREUSED
Product Identity / AI Product VersionAIPSREUSED
Effective Risk Tier / Risk DriversAIPSREUSED
Governance Obligation/ProfileAIPSREUSED
AI Product Assurance / Fitness for PurposeAIPSREUSED / REFERENCED
Agent identity/taxonomyKivaNura Agent OntologyREUSED
Agent Autonomy Level / Control Mode / Tool Use PatternKivaNura Agent OntologyREUSED
Data / Data domain semanticsGEDRM / applicable Data authorityCROSS-DOMAIN-REFERENCE
Data architectureGEDRAARCHITECTURE-REFERENCE only
Information / Analytics semanticsGEARM / applicable authorityCROSS-DOMAIN-REFERENCE
Analytics / Information architectureGEARAARCHITECTURE-REFERENCE only
generic provenanceW3C PROVREUSED / REFERENCED
generic enterprise-AI semantic gaps established M04-M14GEAIRMOWNED
AI reference architecture responsibilitiesfuture GEAIRARESERVED
enterprise target logical realizationHDIPENTERPRISE-SPECIALIZATION; not GEAIRM authority
deployment/operational Product realizationAIPDSRESERVED / REFERENCED

422.1 No authority by namespace proximity​

An AIPS concept appearing in a GEAIRM view remains AIPS-owned.

An Agent Ontology concept appearing beside a GEAIRM-owned relationship remains Agent-Ontology-owned.

A GEDRM/GEARM concept referenced by an AI Context remains externally authoritative.

M17-R15 - Final authority matrix: A1-A15 executable formalization MUST preserve the M17 authority/disposition matrix.

M17-R16 - No shadow authority: Executable GEAIRM artifacts MUST NOT mint shadow copies of externally owned concepts merely to obtain local namespace consistency.


423. Final Cross-Model Alignment​

423.1 BPS / PIN​

BPS answers:

Is it a Product?
What is intrinsically being productized?
What Product Family / Kind applies?
What contextual Product role applies?

GEAIRM answers:

What enterprise-AI concepts and relationships are present?

AIPS then specializes the Product boundary for AI Products.

No GEAIRM assertion bypasses Productness or PIN.

423.2 AIPS​

AIPS is both a key external semantic authority and the specification-family home for GEAIRM.

The final responsibility split is:

AIPS AI Product semantics
When does a Product qualify as an AI Product?
How is that Product Version classified/governed?

GEAIRM
What generic enterprise-AI concepts exist,
and how do those concepts relate irrespective of Product status?

423.3 KivaNura Agent Ontology​

The Agent Ontology owns:

Agent
Agent taxonomy
Agent Autonomy Level
Agent Control Mode
Tool Use Pattern

GEAIRM owns the semantic bridges that relate Agents to AI Execution, authority, Tools, Decision Formation, Environment and broader AI semantics.

423.4 GEDRM / GEDRA​

GEAIRM may consume/reference Data, Data Products and related Data semantics, but Data meaning remains outside GEAIRM authority.

GEDRM and GEAIRM are adjacent semantic reference models; GEDRA and future GEAIRA are adjacent reference architectures.

423.5 GEARM / GEARA​

Analytics/Information and enterprise AI may intersect in recommendation, decision support, evidence and outputs, but GEAIRM does not redefine the complete Analytics/Information domain.

GEARM and GEAIRM are sibling reference models with intersecting relationships.

423.6 W3C PROV​

PROV remains the generic provenance foundation. GEAIRM does not create a shadow provenance ontology.

423.7 GEAIRA​

GEAIRA will consume the stabilized GEAIRM semantics to answer:

How should enterprise AI be architected?

GEAIRA MUST NOT be used retroactively to redefine what GEAIRM concepts mean.

423.8 HDIP​

HDIP remains an enterprise-specific target conceptual/logical platform architecture for Data, Information and Intelligence. It may realize/specialize the generic reference architectures, but its platform organization is not intrinsic GEAIRM semantics.

M17-R17 - Cross-model alignment: Cross-model mappings MUST preserve adjacent-domain authority rather than flatten BPS, AIPS, Agent, Data, Analytics, provenance and architecture concerns into one namespace or hierarchy.


424. Final 0.1 Hard-Distinction Summary​

M15 remains the detailed canonical hard-distinction catalogue. M17 condenses the most important release-level boundaries:

AI != AI Product

AI Primitive != Intelligence Capability
Intelligence Capability != Execution Mode
Execution Mode != Intelligence Effect

Agent != AI Primitive
Agent != AI Product
Agent autonomy != Product AR
autonomy != authority

AI Context != Environment
Retrieval != Grounding
Grounding != correctness
Instruction != Authorization

Sensing != Perception
Control capability != Control Loop
Actuation != PhysicalAction
physical capability != permission

Recommendation != Decision
Decision effect != Decision authority
Decision effect != Authorization
Authorization != Action

Dependency != provenance
Dependency != identity inheritance
Composition != Product
Federation != Topology Scope
Orchestration != AI Product

Productness != AI Product qualification
Model Product != AI Product
AI-enabled Product != AI Product
APT applies only after qualification

Evaluation != Assurance
Evidence != proof
Assurance != Risk Tier
Risk Tier != Governance Obligation

Identity != Version
Lifecycle != Currentness
Currentness != freshness
Supersession != deletion
change != historical mutation

GEAIRM != GEAIRA
semantic relationship != architecture component

M17-R18 - Release-boundary preservation: The M17 release-level hard distinctions MUST be preserved by all A1-A15 machine-readable artifacts and conformance tests.


425. Final Invariant Baseline​

The M15-I01 through M15-I36 invariant catalogue is the normative human-readable invariant baseline for GEAIRM 0.1.

M17 does not add a second competing invariant set.

For formalization:

M15-I01 ... M15-I36
|
v
A3 Semantic Invariant Catalogue
|
+--> machine-executable constraints
+--> hybrid assessment constraints
+--> assessment-only constraints

A3 may add machine-oriented metadata such as severity, validation mode, message, target scope and evidence expectation, but MUST preserve the semantic requirement of each source invariant.

M17-R19 - Invariant-baseline freeze: A3 MUST project the M15-I01 through M15-I36 semantics without weakening, broadening or silently rewriting them.


426. Final Relationship Baseline​

The M15 canonical inventory contains 41 GEAIRM-owned relationship types established across M04-M14 and 15 externally governed relationship/property bridges explicitly consolidated in M15.

M17 freezes this relationship set as the input to A2.

No new relationship is introduced by M17.

The formalization hand-off is:

M15 canonical relationship inventory
|
v
A2 Canonical Relationship Registry
|
+--> authority
+--> preferred label
+--> subject expectations
+--> object expectations
+--> directionality
+--> transitivity
+--> inverse
+--> non-implications
+--> source slice

A2 may mint stable GEAIRM identifiers only for GEAIRM-owned relationships. Externally owned relationships MUST retain their external identity.

M17-R20 - Relationship-baseline freeze: A2 MUST preserve the M15 relationship inventory, authority boundaries and conservative inference behaviour.

M17-R21 - External-identifier preservation: A2/A4-A8 MUST retain authoritative external identifiers for externally owned relationships rather than reminting local substitutes.


427. Human-Readable 0.1 Semantic Baseline Freeze​

With M17 complete, the human-readable GEAIRM 0.1 semantic baseline is frozen for executable formalization.

The freeze includes:

  • semantic authority and jurisdiction rules from M01;
  • purpose/scope/abstraction boundaries from M02;
  • metamodel and modelling rules from M03;
  • canonical concepts and relationships established M04-M14;
  • M15 relationship inventory;
  • M15 hard distinctions;
  • M15-I01 through M15-I36;
  • M16 canonical patterns and counterexamples;
  • M17 D1-D8 domain map;
  • M17 V1-V8 canonical reference views;
  • M17 authority/disposition matrix;
  • M17 cross-model alignment.

427.1 Meaning of freeze​

"Frozen" means:

A1-A15 are expected to project and formalize the approved human-readable semantics rather than redesign them silently.

The freeze does not mean:

  • GEAIRM 0.1 is a final published release;
  • all semantics are machine-executable;
  • all future revisions are prohibited;
  • errors cannot be corrected;
  • governance approval has occurred.

A material semantic change after M17 requires an explicit normative revision/change record and corresponding reconciliation of executable artifacts.

M17-R22 - Freeze semantics: A1-A15 MUST treat M01-M17 as the source semantic baseline and MUST NOT introduce material semantic change without explicit normative revision.

M17-R23 - Release-status restraint: M17 completion MUST NOT be represented as final governed publication or final release promotion.


428. Executable Formalization Hand-Off​

The next programme is the executable formalization sequence:

A1  Canonical Concept Registry
A2 Canonical Relationship Registry
A3 Semantic Invariant Catalogue
A4 RDF/OWL
A5 JSON-LD Context
A6 SHACL
A7 Structural JSON/JSON-LD Schema
A8 GEAIRM Profile / Extension Schema
A9 Canonical Example Set
A10 Positive Test Cases
A11 Negative / Semantic Boundary Test Cases
A12 Automated Conformance Runner
A13 Cross-Artifact Reconciliation
A14 Regression Suite
A15 Release Closure

428.1 A1​

A1 MUST inventory:

  • GEAIRM-owned concepts;
  • reused external concepts;
  • referenced external concepts;
  • Roles;
  • States;
  • Classifications;
  • editorial-only terms that MUST NOT receive machine identifiers.

428.2 A2​

A2 MUST use the M15 relationship inventory and M17 relationship freeze.

428.3 A3​

A3 MUST project M15-I01-I36 and classify validation mode without weakening the semantics.

428.4 A4-A8​

A4-A8 MUST serialize/formalize the model while preserving:

authority
identity
scope
non-equivalence
conservative inference
Product boundaries
historical semantics

428.5 A9-A11​

M16 P01-P15 and C01-C15 provide the initial semantic basis for canonical positive/negative examples and test cases.

428.6 A12-A14​

Machine PASS MUST validate only what the machine artifacts can validly test. Assessment-only semantics MUST NOT be falsely reported as automatically proven.

428.7 A15​

A15 may close the 0.1 executable candidate only after cross-artifact reconciliation and regression checks succeed. Final governed publication remains a separate promotion decision.

M17-R24 - Formalization fidelity: A1-A15 MUST formalize the M01-M17 semantics and MUST NOT substitute implementation convenience for semantic fidelity.

M17-R25 - Machine-conformance restraint: Automated conformance MUST NOT be represented as proof of semantics that remain hybrid-assessment or assessment-only.


429. Final Slice Completion Matrix​

SliceSubjectM17 status
M01Semantic Authority & JurisdictionCOMPLETE / retained
M02Purpose, Scope & PositioningCOMPLETE / retained
M03Reference-Model Metamodel & Modelling RulesCOMPLETE / retained
M04Core AI & Intelligence SemanticsCOMPLETE / retained
M05AI Primitive & Intelligence Mechanism ModelCOMPLETE / retained
M06Intelligence Capability, Execution & Effect ModelCOMPLETE / retained
M07Data, Knowledge, Context & Grounding ModelCOMPLETE / retained
M08Agent, Agency, Autonomy & Tool ModelCOMPLETE / retained
M09Embodiment, Environment & Control-Loop ModelCOMPLETE / retained
M10Decision, Recommendation, Plan & Action SemanticsCOMPLETE / retained
M11Dependency, Composition & Topology ModelCOMPLETE / retained
M12Product Boundary & AIPS BridgeCOMPLETE / retained
M13Evaluation, Assurance, Risk & Governance ModelCOMPLETE / retained
M14Identity, Version, Lifecycle, Currentness & ProvenanceCOMPLETE / retained
M15Canonical Relationship Model & Hard DistinctionsCOMPLETE / retained
M16Canonical Patterns, Examples & CounterexamplesCOMPLETE / retained
M17Reference Views, Alignment & 0.1 ConsolidationCOMPLETE

No earlier slice is superseded by M17. M17 consolidates them.

M17-R26 - Slice preservation: M17 completion MUST NOT be interpreted as replacing or nullifying normative rules from M01-M16.


430. GEAIRM 0.1 Consolidated Definition​

The final human-readable definition is:

GEAIRM 0.1 is a technology-neutral, mechanism-neutral enterprise Artificial Intelligence reference model that identifies, organizes and relates the generic concepts needed to describe Artificial Intelligence mechanisms, capabilities, executions, effects, context, grounding, Agents, authority, decisions, actions, environment interaction, Products, dependencies, evaluation, assurance, governance, identity and evolution while preserving the semantic authority of BPS, AIPS, the KivaNura Agent Ontology, GEDRM/GEARM, W3C PROV and other applicable domain authorities.

GEAIRM is therefore:

a semantic reference model
not a Product qualification engine
not an AI deployment specification
not a reference architecture
not an implementation architecture
not a universal replacement ontology

Its core modelling philosophy is:

Reuse authoritative meaning.
Mint only genuine semantic gaps.
Keep orthogonal concerns orthogonal.
Do not infer Productness from AI participation.
Do not infer authority from autonomy.
Do not infer architecture from semantics.
Do not rewrite history when currentness changes.

M17-R27 - Final-definition fidelity: Subsequent machine-readable artifacts and GEAIRA work MUST remain consistent with the consolidated GEAIRM 0.1 definition.


431. M17 Normative Rules​

RuleRequirement
M17-R01M17 MUST consolidate M01-M16 without silently creating new semantic meaning.
M17-R02D1-D8 are organizational semantic views and MUST NOT become intrinsic ontology types by view membership.
M17-R03Cross-domain appearance MUST preserve one governed element identity unless independently reclassified.
M17-R04All domains/views MUST preserve authority, Product-boundary, conservative-inference and historical-preservation rules.
M17-R05The V1 authority landscape MUST govern 0.1 formalization unless normatively revised.
M17-R06Mechanism, capability, execution mode and effect MUST remain orthogonal in V2/formalization.
M17-R07Context, retrieval, grounding, memory, instruction and Data/Information authority MUST remain distinct.
M17-R08Agent identity, Agent autonomy/control, Product AR, Delegated Authority and Authorization MUST remain distinct.
M17-R09Recommendation, Plan, Decision Formation, Decision effect, Authorization and Action MUST remain independently representable.
M17-R10Environment interaction, Embodiment, Agent identity, Control, Actuation, PhysicalAction and authority MUST remain distinct.
M17-R11Dependency/Composition/Federation/Orchestration MUST NOT create Product or AI Product identity absent BPS/AIPS qualification.
M17-R12APT, Derivation Topology and Topology Scope MUST remain AIPS-governed Product classifications.
M17-R13Evaluation, Evidence, Assurance, Fitness, Risk and Governance MUST remain separate semantic axes.
M17-R14Identity/version separation, currentness contextuality, historical non-mutation, provenance/dependency distinction and scoped supersession MUST be preserved.
M17-R15A1-A15 MUST preserve the final M17 authority/disposition matrix.
M17-R16Executable GEAIRM artifacts MUST NOT create shadow copies of externally owned concepts for namespace convenience.
M17-R17Cross-model mappings MUST preserve adjacent-domain authority and MUST NOT flatten all concerns into one hierarchy.
M17-R18M17 release-level hard distinctions MUST be preserved by A1-A15 artifacts and tests.
M17-R19A3 MUST project M15-I01-I36 without semantic weakening or silent broadening.
M17-R20A2 MUST preserve the M15 relationship inventory, ownership and conservative inference.
M17-R21A2/A4-A8 MUST retain authoritative external identifiers for externally owned relationships.
M17-R22A1-A15 MUST treat M01-M17 as the source semantic baseline and require explicit revision for material semantic change.
M17-R23M17 completion MUST NOT be represented as final governed publication or release promotion.
M17-R24A1-A15 MUST prefer semantic fidelity over serialization or implementation convenience.
M17-R25Automated conformance MUST NOT claim proof of hybrid-assessment or assessment-only semantics.
M17-R26M17 consolidation MUST NOT nullify or replace normative rules from M01-M16.
M17-R27Machine artifacts and future GEAIRA work MUST remain consistent with the consolidated GEAIRM definition.
M17-R28M17 introduces no new GEAIRM-owned relationship type.
M17-R29M17 introduces no new Product, Agent, risk, lifecycle or architecture taxonomy.
M17-R30M17 domain/view labels are documentation constructs and MUST NOT receive executable ontology identity unless a later normative revision explicitly promotes them.
M17-R31GEAIRM executable formalization MUST preserve M15 non-equivalence vs formal-disjointness restraint.
M17-R32Cross-domain Data/Information references MUST preserve GEDRM/GEARM semantic authority and MUST NOT be locally redefined merely for AI convenience.
M17-R33PROV remains the generic provenance authority; GEAIRM MUST NOT create a shadow provenance hierarchy during formalization.
M17-R34GEAIRA MUST consume stabilized GEAIRM semantics and MUST NOT retroactively redefine GEAIRM concept meaning through architecture design.

M17 establishes 34 normative rules, M17-R01 through M17-R34.


432. M17 Exit Criteria​

GEAIRM-M17 is complete when:

  • the final D1-D8 semantic-domain map is established as a view structure rather than ontology taxonomy;
  • V1-V8 canonical reference views are established;
  • BPS, AIPS, Agent Ontology, GEDRM/GEDRA, GEARM/GEARA and PROV authority boundaries are explicitly reconciled;
  • Product, Agent, Data, Information/Analytics, provenance and architecture concerns retain their governing authorities;
  • the M15 hard-distinction catalogue remains authoritative;
  • M15-I01 through M15-I36 remain the invariant baseline;
  • the 41 GEAIRM-owned relationship baseline and 15 external relationship bridges remain frozen for A2;
  • M16 P01-P15 / C01-C15 remain the canonical human-readable scenario baseline for A9-A11;
  • no new domain concept or relationship is created by M17 consolidation;
  • the human-readable M01-M17 semantic baseline is explicitly frozen for formalization;
  • freeze is distinguished from final governed publication;
  • A1-A15 hand-off rules are explicit;
  • machine conformance is prevented from overclaiming assessment-only semantics;
  • future GEAIRA responsibility is explicitly downstream of stabilized GEAIRM meaning;
  • M01-M16 normative rules remain in force.

M17 conclusion: these criteria are satisfied. The human-readable GEAIRM 0.1 semantic baseline is now complete and frozen for executable formalization. The next programme step is GEAIRM-A1 - Canonical Concept Registry.


433. Forward Work​

The next slice/programme step is:

GEAIRM-A1 - Canonical Concept Registry

A1 will project the M01-M17 human-readable semantic baseline into the first machine-oriented canonical registry. It will assign stable GEAIRM identifiers only to GEAIRM-owned canonical elements, preserve authoritative external identifiers for reused concepts, record M03 element kinds and authority dispositions, and explicitly retain editorial-only terms that MUST NOT become ontology entities.